Professional Documents
Culture Documents
Kuznetsova J Linguistic Profiles Going From Form To Meaning
Kuznetsova J Linguistic Profiles Going From Form To Meaning
Kuznetsova J Linguistic Profiles Going From Form To Meaning
Linguistic Profiles
Cognitive Linguistics Research
Editors
Dirk Geeraerts
John R. Taylor
Honorary editors
René Dirven
Ronald W. Langacker
Volume 53
Julia Kuznetsova
Linguistic Profiles
DE GRUYTER
MOUTON
ISBN 978-3-11-035553-6
e-ISBN (PDF) 978-3-11-036185-8
e-ISBN (EPUB) 978-3-11-039348-4
ISSN 1861-4132
www.degruyter.com
Acknowledgements
This book builds upon my dissertation “Linguistic Profiles: Correlations be-
tween Form and Meaning” that I defended in 2013 at the Arctic University of
Norway (Tromsø). I am deeply indebted to my supervisor, Laura Janda, for all
her support, help, and non-stop optimism throughout this project. I am also
grateful to my fellow members of the CLEAR group at the University of Tromsø:
Tore Nesset, Olga Lyashevskaya, Anastasia Makarova, Svetlana Sokolova, and
Anna Endresen, for their interest in my work, encouragement, and input. Parts
of my dissertation were researched and written during the 2011/2012 academic
year when, together with other members of the CLEAR group, I was part of the
research group at the Centre for Advanced Study (CAS) at the Norwegian Acad-
emy of Science and Letters in Oslo working on the project “Time is Space: Un-
conscious Models and Conscious Acts”. I am thankful to all the other members
of the CAS group for their comments and collaboration.
I am deeply indebted to my dissertation opponents, Mark Turner from Case
Western Reserve University and Maria Polinsky from Harvard University, who
helped me to improve the text with their very useful comments. I am also grate-
ful to Dirk Geeraerts from the University of Leuven. As an editor to the series, he
provided very insightful suggestions that were a tremendous help to me while I
was writing this book. I would also like to say thank you to members of the
Moscow and Saint-Petersburg linguistic community for their thoughtful com-
ments and advice.
I would like to express my gratitude to the University of Tromsø and the
Centre for Advanced Study in Oslo for providing friendly and inspiring academ-
ic environments. I would like to thank also the University of Chapel Hill in
North Carolina for providing me with working space and access to its library
while I was working on this book. I would like to thank Mary K. Brown for her
scrupulous copy-editing. I am also grateful to Birgit Sievert and Julie Miess at
De Gruyter Mouton publishing house for their advice and help throughout this
work. Finally, I wish to thank my friend Antonina Manucharova, my husband
Alexander Rubin, and my sister Eugenia Kuznetsova for their constant support.
Table of content
Note on corpora, statistical methods and databases | xi
1. Introduction | 1
2. Linguistic profiles | 4
2.1. Form and meaning in cognitive and functional linguistics | 4
2.2. Multilevel profiles | 14
2.2.1. Behavioral profiles | 15
2.2.2. Dangers of multilevel profiling | 16
2.3. Monolevel profiles | 27
2.3.1. Constructional profiles | 27
2.3.2. Grammatical profiles | 27
2.3.3. Semantic profiles | 28
2.3.4. Radial category profiles | 29
2.4. Summary | 29
6.2.4. V telefon ‘into the phone’ and v trubku ‘into the receiver’ | 188
6.3. Why do we scream into the phone? | 190
6.4. Interaction of the possessive construction and the construction of disap-
pearance | 194
6.5. Summary | 200
7. Conclusion | 202
References | 208
Appendix | 221
structure constructions available for an imperfective verb and its correlate per-
fective verb, producing an objective measure of interchangeability between the
two verbs.
Chapter 6 examines the relationship between a construction and its parts
from another point of view. This chapter explores the relationship between
constructions and the fillers of a constructional slot. I offer a new method, col-
lostructional profiling, which shows that a construction has strong semantic
requirements for its slots and show that these requirements can be seen and
deduced from the list of frequent fillers of that slot.
Chapter 7 summarizes the findings discussed in the book, arguing that form
and meaning are in a relation of statistical correlation and that this correlation
allows us to investigate the distribution of form in order to deduce the distribu-
tion of meaning.
2. Linguistic profiles
2.1. Form and meaning in cognitive and functional linguistics
The relationship between the form and meaning of a linguistic unit and lin-
guists’ understanding of that relationship has evolved throughout the history of
linguistics. Within cognitive and functional linguistics, the history of this rela-
tionship departs from critiques of generative grammar, where form was claimed
to be “autonomous and independent of meaning” (Chomsky 1957: 17). Chom-
sky’s postulate of the independence of syntax and semantics was first ques-
tioned in the generative semantics program, which already in 1967 had stated
that “[s]yntactic and semantic representations are related via transformations”
(Huck and Goldsmith 1995: 20).
This idea was developed even further in the 1970s and 1980s when several
approaches emerged that concentrated on connections between syntax and
semantics. These approaches were developed to a large extent as an answer to
the Chomskian generative framework, and some of the founders of those ap-
proaches came from among the ranks of the generativists; for example, George
Lakoff was one of the initiators of the generative semantics research program.
The most significant of these newer approaches are cognitive linguistics and
functional linguistics. The former grew out of works of Charles Fillmore, George
Lakoff, Ronald Langacker, and Leonard Talmy, and the latter is associated with
linguists such as Joan Bybee, Bernard Comrie, John Haiman, Paul Hopper, San-
dra Thompson, and Talmy Givón.
Whereas the main goals of cognitive linguistics are “to examine the relation
of language structure to things outside language: cognitive principles and
mechanisms not specific to language, including principles of human categoriza-
tion; pragmatic and interactional principles; and functional principles in gen-
eral, such as iconicity and economy” (Kemmer 2010), functional linguistics
focuses on “explanatory principles that derive from language as a communica-
tive system, whether or not these directly relate to the structure of the mind”
(ibid.). Thus, despite slightly different foci, cognitive linguistics and functional
linguistics agree on their view of the relationship between form and meaning,
which can be summarized by this quote from Haspelmath (2002: 2): “[T]he ex-
planation of language form involves appreciating the regularities of language
function.”
Form and meaning in cognitive and functional linguistics | 5
In formal grammar, the language phenomenon is the means of description, the material
on which arguments for the construction of the model are based… Functional grammar
broadens its purview. It too analyzes grammatical structure. But it also analyzes: the
communicative situation, the purpose of the speech event, the participants, and the dis-
course context.
Because the cognitive and functional approaches are so close, some recent stud-
ies unite them using the term “cognitive-functional linguistics” (see Tomasello
2003: 5). Cognitive and functional linguistics introduced three new ideas to the
relationship of form and meaning. First, language is meaning-driven. Second,
every distinction in a language is gradual. Third, linguistics is usage-based.
The first idea states that “grammar is meaningful” (Langacker 2008: 3) and
that meaning is the central driving force of every linguistic distribution.
Grammatical markings and patterns call attention to subtle aspects of meaning and pose
descriptive problems requiring semantic solutions. In doing semantic analysis a linguist
can use these as both a stimulus and a check: besides being psychologically plausible and
internally well motivated, semantic descriptions must articulate well with grammar (ibid.:
14).
Table 1: Examples of constructions varying in size and complexity; form and function are speci-
fied if not readily transparent (from Goldberg 2003)
Covariational Conditional Form: The Xer the Yer Meaning: linked independent
construction (e.g. The more you think about and dependent variables
it,
the less you understand)
Ditransitive (double object) Form: Subj [V Obj1 Obj2] Meaning: transfer (intended or
construction (e.g. He gave her a Coke; actual)
He baked her a muffin.)
Passive Form: Subj aux VPpp (PPby) Discourse function: to make
(e.g. The armadillo was hit undergoer topical and/or actor
by a car) non-topical
Today, the definition of a construction has been extended, and it is now claimed
that “[i]n addition, patterns are stored as constructions even if they are fully
predictable as long as they occur with sufficient frequency” (Goldberg 2006: 5).
This addition recognizes the relationship between constructions in grammar
and constructions in the mind. With regard to grammar, it is reasonable to de-
scribe as constructions only those form/meaning pairs where the whole cannot
be seen as the sum of the parts. Yet recent research shows that language acqui-
sition and language use are better described if we acknowledge that frequent
patterns are stored in the minds of speakers as “prefabricated units” to ease
access to them (see Dąbrowska 2004).
Summing up, Construction Grammar presents a point of view that departs
from generative grammar. Although generative grammar has excluded meaning
from consideration and made regularity of syntactic form a starting point for
research, linguists who work within the framework of Construction Grammar
insist that meaning is present at every level of language. Furthermore, they
claim that such meaning tends to be somewhat arbitrary, or noncompositional.
At the same time, arbitrariness between form and meaning does not presuppose
that the two are not connected. On the contrary, meaning influences form, and
differences in form are always correlated with differences in meaning. All lin-
guistic profiles discussed in this book take Construction Grammar as their theo-
retical foundation, i.e. the recognition that a construction is the basic unit of
language.
Other main ideas of cognitive-functional linguistics that have transformed
the relationship of form and meaning are prototypicality and gradient catego-
ries. Prototypicality is the assumption that referents of the same form are struc-
tured in terms of meaning. There are “best” category members (prototypes) and
members that systematically deviate from the “best case” (peripheral). The
cognitive approach replaced many clear-cut boundaries in linguistics by the use
of gradient scales. Bybee (2010: 6) argues that gradience in language is a result
of language being in constant use: “If language were a fixed mental structure, it
would perhaps have discrete categories; but since it is a mental structure that is
in constant use and filtered through processing activities that change it, there is
variation and gradation.” There are difficulties in defining such basic linguistic
ideas as words, grammatical categories, syntax vs. lexicon, or grammatical vs.
ungrammatical. All these concepts appear in so many guises that defining them
is much easier in terms of prototypes and periphery.
8 | Linguistic profiles
The idea of prototypes was first introduced in the pioneer work of Rosch
(1973, 1975) who showed that a set of signifieds denoted by a word produces a
radial category. For example, the word bird in English has birds such as a robin
as a prototype, and less prototypical birds such as an ostrich or penguin as
periphery. Gradience in the domain of linguistic categories is discussed in the
framework of Fuzzy Grammar (Lakoff 1973, Mohan 1977, Quirk 1965, Ross 1972,
1973a, 1973b) and a related approach in sociolinguistics called the Variable
Rules framework (Cedergren and Sankoff 1974, Labov 1969). The optimal gram-
matical description in these approaches includes characterization of the degree
of departure from the prototype. Also important for the issue of form and mean-
ing in this approach is the structure of the periphery. Categories do not have
distinct boundaries; the edges of the categories are “fuzzy”. This statement can
be illustrated with the example of the so-called “fake NP squish” – the argu-
ment introduced by Ross (1973a) to show that the grammatical category of noun
phrase (NP) does not have clear boundaries. If we consider the list in (1), each
subsequent item in it is less prototypical in terms of “being an NP”: animates
show the highest degree of NP status; for example, they easily allow reapplica-
tion of the rule of raising, whereas this rule is problematic for events and not
possible for other types of NPs presented in (1); see (2).
(1) a. Animates
b. Events
c. Abstracts
d. Expletive it
e. Expletive there
f. Opaque idiom chunks
It is sometimes maintained that syntax and lexicon are sharply distinct and quite dissimi-
lar in nature. Syntax, on this view, is characterized by regularity and is thus describable
by rules, whereas lexicon is the repository of irregularity—a mass of unpredictable idio-
syncrasies. CG takes a very different stand on these issues. It offers a unified account of
lexicon and grammar (subsuming both syntax and morphology), in which every linguistic
unit represents a reinforced pattern and thus embodies a regularity of some kind.
Form and meaning in cognitive and functional linguistics | 9
The problem of language acquisition has often been summarized in terms of the problem
of the poverty of the stimulus. Our linguistic capacity, for instance our knowledge of
English, goes beyond the evidence we have been exposed to in our childhood. The linguist
wants to account for the fact that the linguistic competence is attained in spite of im-
portant inadequacies in the stimulus, the linguistic experience. Three types of inadequa-
cies are standardly referred to in the literature. First, we do not just come across grammat-
ical sentences: everyday use of language contains slips of the tongue, hesitations,
incomplete sentences, etc. Second, the experience, i.e. the stimulus, is finite, and we end
up being able to produce and process an infinite number of sentences. Third, we acquire
knowledge about our language for which we have no overt or positive evidence in our
experience (bold marked by Haegeman).
have led to this state of affairs in linguistics. Geeraerts compares the work of a
modern semanticist with the work of a doctor. The doctor, when presented with
a patient, produces several hypotheses as to the diseases that may have caused
the observed condition. Then, the doctor conducts tests, such as physical tests,
blood tests, and medical imaging, and using the results of the tests can choose
among the hypotheses. Here, the doctor uses the combined experience of his or
her predecessors in accordance with which certain testable symptoms correlate
with one or the other interpretation of the patient’s condition.
Geeraerts argues that the analysis conducted by a semanticist is in some
ways similar to the analysis conducted by a doctor. However, for a long time
semanticists relied only on the first step of the diagnostic procedure: intuitive
insight. Geeraerts (ibid.: 64) sees two main “factors that prevent linguistic se-
mantics from embracing the empirical method more frequently and enthusiasti-
cally than it actually does.” First, it may seem that semantics can scarcely be
studied quantitatively: “[H]ow can meaning, the most qualitative of all linguis-
tic features, be expressed in numbers, and more broadly, how could meaning,
the most ephemeral and subjective of all linguistic phenomena, be tackled with
methods that aim at objectivity?” (ibid.). Second, one might argue that seman-
tics does not need any method other than introspection. Geeraerts answers the
first question by comparing linguistics with psychology. If we believe that the
ability to speak a language is cognitive, then we must agree that linguistic
meaning may be subjected to controlled experimentation and quantitative
analysis of data in the same way that experimental psychology investigates
cognitive phenomena at large. In answering the second question, Geeraerts
reasons that no one is an ideal speaker of a language in the Chomskian sense:
(3) Look, if people want to discuss what the pope said about the death pen-
alty, they ought read his encyclical "Evangelium Vitae”. [Jesse Jackson
Joins Parents of Terri Schiavo (2005) CNN_Crossfire]
(4) We should to look at them and think about what they're thinking about
their own failures1. [The roundtable; this week’s politics (2010. ABC This
Week]
||
1 Although examples of the verb ought without to and examples of the verb should with to are
attested in the corpus, native speakers of English find sentences such as (3) more acceptable
than sentences such as (4).
Form and meaning in cognitive and functional linguistics | 13
Then the question arises: How do we deal with such phenomena? How can we
answer whether these counterexamples to our hypothesis can be discarded?
Such questions have led linguists, like many social scientists, to the use of sta-
tistical methods. For example, for the case of the verbs ought and should, Rad-
ford’s rules can be easily reformulated to describe actual tendencies. Table 2,
summarizing the data from COCA, shows us that the verb ought strongly prefers
complements with to (99.85% of all uses), whereas the verb should almost al-
ways is used without to (99.99% of all uses). A Fisher test gives the
P-value < 2.2e-16, which indicates that this distribution could not occur by
chance.
Table 2: Distribution of complements with the modal verbs ought and should
to+V V
ought 17,521 27
should 16 210,122
One could say that the use of quantitative methods for corpus data is where
cognitive and functional linguistics were destined to arrive. Indeed, once lin-
guistic phenomena began to be studied on the basis of corpus data and once the
fact was accepted that the linguistic phenomena are gradual, the only method
for testing hypotheses that remained was statistical analysis. Statistics allows
one to measure whether an observed tendency is significant and to evaluate the
effect size of counterexamples to the tendency.
Linguists have long struggled to make linguistics an empirical science. The
descriptivists who endeavored to make the rules of description stricter made an
important early attempt. Another attempt was made within the Chomskian ap-
proach, which tried to reach the same goal by formalizing description. Early
cognitivists and functionalists, however, believed that grammar is meaningful,
but that this approach “offered no means for testing hypotheses or falsifying
analyses,” as Glynn (2010: 2) pointed out. Now, we can apply statistical meth-
ods to corpus distributions and show that the distribution of form is driven by
meaning. So finally, we have the tools to prove that form is actually meaningful.
A corpus-driven quantitative approach to semantics allows linguistics to be-
come a truly empirical science, where linguists can study language “in its natu-
ral habitat” (Dąbrowska 2004: 2) and test how well the proposed theories assess
the actual data.
Summing up, the cognitive and functional approaches to linguistics added
several new ideas to the relationship between form and meaning. First, mean-
ing is connected with syntactic form, and thus, semantics and pragmatics influ-
14 | Linguistic profiles
This and next sections discuss linguistic profiles – several approaches that are
currently being pursued within the field of cognitive linguistics. The various
types of linguistic profiles are based on Construction Grammar, i.e. they recog-
nize the construction as the basic element of language. The main idea that un-
derlies linguistic profiling is that a strong correlation exists between semantic
and distributional properties. This idea is not new in linguistics and is described
at length in the scholarly literature. The well-known quote from Firth (1957: 11)
is: “You know a word by the company it keeps.” The same idea is stated more
formally by Harris (1970: 758):
The same assumption is made in the paper by Gries and Otani (2010: 122): “[T]he
distributional characteristics of the use of an item reveal many of its semantic
and functional properties and purposes.”
If form and meaning are closely related, we can explore the distribution of
form and, as a result, obtain information about meaning. Janda and Lyash-
evskaya (2013: 3) describe linguistic profiles as a group “of related methodolo-
gies for probing the statistical behavior of linguistic units.” Researchers who
develop linguistic profiles investigate meaning using form as a tool. There are
various ways to do this. Using grammatical profiling (Janda and Lyashevskaya
2011c), we can analyze how different inflections are distributed, depending on
the semantics of the lexeme. Using semantic profiling (Janda and Lyashevskaya
2013), we can move from inflection to derivation and investigate how different
Multilevel profiles | 15
affixes or different submeanings of one affix are chosen, given the meaning of
the base. Via constructional profiling (Janda and Solovyev 2009, Sokolova,
Lyashevskaya and Janda 2012), we can examine which constructions are chosen
by a lexeme and how that choice depends on the meaning of the lexeme. Radial
category profiling (Nesset et al. 2011, Endresen et al. 2012) allows us to compare
the radial networks of related linguistic units. Finally, I propose a method that I
call collostructional profiling, in which we can characterize one slot in a con-
struction and find its semantic preferences by looking at frequent fillers of that
slot. These methods are all inspired by behavioral profiling, which analyzes
how a variety of grammatical, syntactic, lexical, and semantic features group
together and form a cluster of linguistic units (Divjak and Gries 2006, Gries and
Divjak 2009, Gries and Otani 2010).
||
2 Near synonyms are words that are close in meaning, but not exactly synonymous. For exam-
ple, try and endeavor both mean ‘to make an effort to accomplish’; however, these two words
are not always interchangeable, as in try a new method vs. *endeavor a new method.
16 | Linguistic profiles
all the variables are chosen by the researcher, so the study has a high chance of
finding that a certain included variable is either important or not important for
the clustering, but there might exist an additional important variable not in-
cluded in the set, and thus, the behavioral approach would not let us see the
impact of the missing variable. In order to avoid this problem we would have to
investigate all available variables of a certain level. This task becomes easier
once the field of investigation is narrowed. For example, we can concentrate on
morphological tagging and explore all possible inflections of the verb for tense
and aspect.
Another advantage can be gained if we narrow the scope of our analysis to a
single linguistic level is that we can study more data. When we study behavioral
profiles we need to include many variables in our analyses; so, even nine mem-
bers of a semantic field result in a long tagging process. Therefore, behavioral
profiles are most often studied using several close synonyms only. Focusing
exclusively on one level allows us to increase the number of studied lexemes.
See, for example, studies on grammatical profiling that can include all verbs in
a corpus, as in the study by Janda and Lyashevskaya (2011c) as well as the
grammatical profiling study described in chapter 3.
Another problem with behavioral profiling is that variables belong to differ-
ent levels (morphology, syntax, and semantics). The following section discusses
this problem and the issues that arise from this approach using two case stud-
ies.
One may ask why we need to confine ourselves to only one level. Can we not
simply tag everything that is available to us and then feed it all into a statistical
software package and let the program “pick the winners”? The danger that
awaits us on this path is multicollinearity, a condition in which independent
variables used for prediction correlate with each other.
Language variables rarely have random distribution. Morphology is related
to syntax, syntax is related to semantics, semantics is related to morphology,
and all of these variables are related to frequency. Thus, one of the variables in
a multilevel profile is by necessity correlated to another variable within that
profile. For example, the distribution of case endings is correlated with anima-
cy, the distribution of semantic classes is correlated with frequency, the distri-
bution of syntactic constructions is correlated with semantic classes, and so on.
Therefore, using variables that belong to morphology, syntax, semantics, and
frequency in one model can lead to multicollinearity.
Multilevel profiles | 17
These case studies use semantic classes of verbs as tagged in the RNC. The
classification follows the general ideas of the Moscow Semantic School3. All
semantic tags that are pertinent to these case studies are shown in Table 3.
Table 3: Semantic tags of imperfective verbs that form perfectives with prefixes s- and za-
||
3 The classification is based on works of the Leksikograf group, cf. Padučeva 2004, Kustova et
al. 2005, and more generally Apresjan 2005. Details regarding this classification are available
at: http://www.ruscorpora.ru/en/corporasem.html.
Multilevel profiles | 19
Table 4: Significance of the logistic regression models predicting the choice between prefixes
s- and za-
χ2 degrees of freedom P
grammatical model 13.7 5 0.008
semantic model 71.9 17 4.7e-09
frequency model 30.2 2 3.9e-08
combined model 92.8 22 5.3e-11
!
20 | Linguistic profiles
Table 5: P-values of proportions of the grammatical forms and the frequency of the imperfec-
tive verb in separate models and in the combined model
Table 6: Intransitive imperfective verbs that form perfectives via prefixes za- and s- (morpho-
logical class 1a, with frequency between 10 and 300 occurrences in the RNC)
these verbs have the same tag, i.e. ‘change of state’, whereas the distance be-
tween the verb za-gustet’ ‘become dense’ and za-beremenet’ ‘get pregnant’ is 1,
because the former has the tag ‘change of state’ and the latter has the tag ‘phys-
iological'.
Finally, grammatical distance measures the difference between the gram-
matical profiles of the verbs. The distance between grammatical profiles is de-
fined as the square root of the sum of the squares of difference between the
proportions of the present, past, infinitive, and imperative forms, respectively.
For example, the proportions for the verb gustet’ ‘become dense’ are 0.57 for
present tense, 0.38 for past tense, 0.02 for infinitive, and 0.03 for imperative,
and the proportions for the verb kosnet’ ‘stagnate’ are 0.65 for present tense,
0.17 for past tense, 0 for infinitive, and 0.17 for imperative. Therefore, the
grammatical distance between these two verbs is
According to this measure, verbs that have similar grammatical profiles have a
smaller grammatical distance between them, whereas verbs with different
grammatical profiles have a larger grammatical distance between them.
Finally, I propose two combined distances. The first combined distance is
the sum of the grammatical distance and the semantic distance. It indicates
whether the two verbs have similar grammatical profiles and the same semantic
tag. The second combined distance is the sum of the grammatical distance, the
prefix distance, and the semantic distance, and reflects whether the two verbs
are close in terms of their grammatical profiles, have the same prefix, and be-
long to the same semantic class.
Figure 1 shows a cluster based on grammatical distance. We can see that
although this cluster is based on grammatical profiles, the verbs in each two-
verb cluster share the same prefix. Except for two three-verb clusters, the verbs
in the three-verb clusters also share the same prefix.!The rightmost cluster com-
bines s-ozorničat’ ‘act outrageously once’ with two verbs that have the prefix
za-: za-gustet’ ‘get dense’ and za-mlet’ ‘get thrilled with’. The cluster in the
middle combines s-podličat’ ‘act meanly’ with two verbs with the prefix za-: za-
xiret’ ‘decay’ and za-beremenet’ ‘become pregnant’. Many of the verbs within
the final two-verb subclusters and three-verb subclusters also share the same
semantic class. This grouping occurs because grammatical profiles and prefixes
are strongly correlated with the verbs’ meanings.
Multilevel profiles | 23
Figure 2 presents a cluster built using the first combined distance. This cluster
groups together all the verbs that share both a grammatical profile and a se-
mantic tag. Note also that the verbs are divided into two large subclusters; the
first contains only verbs with the prefix s-, and the second contains only verbs
with the prefix za-. This division happens because the semantic tags correlate
with the verbal prefixes in these groups. Verbs of behavior are attracted to the
prefix s-, whereas verbs of change of state and physiological verbs are attracted
to the prefix za-.
24 | Linguistic profiles
Fig. 2: Cluster of fourteen verbs based on grammatical profiles and semantic classes (using
first combined distance)
Figure 3 shows the cluster that is based on the second combined distance. This
cluster groups together verbs that have similar grammatical profiles, have the
same semantic class, and have the same prefix. Interestingly, the clusters
shown in Figure 2 and Figure 3 have similar structures. For example, the right-
most verb in Figure 2, s”-exidničat’ ‘say something malicious’, is combined in a
two-verb cluster with the verb s-žul’ničat’ ‘cheat’. The same two-verb cluster is
found in the rightmost part of Figure 3. In Figure 2, this two-verb cluster com-
bines with a five-verb cluster that contains the remaining verb with the prefix s-.
A similar picture is found in Figure 3. All the other verbs in Figures 2 and 3 also
Multilevel profiles | 25
appear in exactly the same clusters. The only difference between Figures 2 and 3
is the height of the dendrogram. The dendrogram in Figure 2 starts branching at
the height of 1.5, whereas the dendrogram in Figure 3 starts branching at the
height of 2.5; that is, the same clusters now appear lower in the dendrogram.
Fig. 3: Cluster of fourteen verbs based on grammatical profiles, semantic classes and prefixes
The model that is based on the first combined distance contains grammatical
and semantic information. The model based on the second combined distance
has, in addition to the first model’s information, information about prefix use.
However, this new information does not affect the way that the verbs in the
cluster are grouped. The grouping of the verbs does not change when infor-
mation about prefixes is added, because in this case study specifically chosen
26 | Linguistic profiles
ers. A more focused type of profiling makes it possible to take into account all
possible variables at a given level and, therefore, not to miss any of the varia-
bles that could affect a distribution. Examples of more focused linguistic pro-
files proposed in the literature are discussed in the following sections.
One way to solve the problems associated with behavioral profiling is to employ
constructional profiling (Janda and Solovyev 2009). In this approach, we can
explore different constructions in which the lexeme of interest can be used. This
approach is based on the following assumption:
If one accepts the claim of construction grammar that the construction is the relevant unit
of linguistic analysis, then we should expect to find a relationship between the meanings
of words and the constructions they are found in (ibid.: 367).
The authors test their approach on a case study of Russian aspectual pairs and
show that suffixation and prefixation yield imperfective/perfective verb pairs
that show similar behavior in terms of grammatical profiles, supporting the
hypothesis of the parallelism between these two types of aspectual derivation.
The use of grammatical profiles, like the use of constructional profiles,
solves the two problems of behavioral profiles: overlapping variables and de-
pendence on the researcher. Grammatical profiling, like constructional profil-
ing, operates using elements from a single level, in this case the morphological
level, and is independent of the researcher’s presuppositions. Thus, this method
provides a useful resource for exploring differences in meaning by investigating
differences in form. In chapter 3, grammatical profiling is used to study the
distribution of gender markings of Russian verbs and how this distribution
characterizes the semantics of a verb.
2.4. Summary
The relationship between form and meaning has a long history in the field of
linguistics. Starting from stating complete arbitrariness between form and
meaning at the level of a word, and then going through a period of willful disre-
gard of meaning, linguists found themselves with two modern approaches to
the relationship between syntactic form and meaning. One of these approaches
30 | Linguistic profiles
states that syntax is autonomous from meaning, and the other recognizes a
strong correlation between form and meaning and offers several methodologies
that allow us to explore meaning based on the distribution of form.
Cognitive and functional linguistics introduced three new ideas regarding
the relationship between form and meaning. The first of these ideas is that form
is meaningful, i.e. the distribution of form is related to the distribution of mean-
ing. The second idea proposes that meaning is gradient and is best described
not in terms of clear-cut boundaries, but in terms of scales and radial categories.
Because linguists found themselves facing gradual phenomena, they have
adopted tools designed for dealing with gradual phenomena, that is, statistical
methods. Last but not least, language is usage-based, and therefore, everything
a learner of a language needs can be extracted from language use. As a result,
the focus of linguistics has shifted to the corpus as an available source of real
language use.
The methods presented in this book demonstrate a strong relationship be-
tween form and meaning and allow us to establish correlations between mean-
ing and distribution of form. Such correlations serve as an additional argument
for the concept of “meaningful grammar.” The correlations show that meaning
underlies the distribution of forms.
3. Grammatical profiling and gender stereotypes
In Russian the past tense form of a verb contains a gender marker agreeing with
the subject. This gives us a unique opportunity to explore the verbs’ distribution
by gender, using the relation between forms with feminine and masculine end-
ings as the basis for a feminine:masculine ratio. The present study shows that
an average Russian verb in the Russian National Corpus (RNC) has three mascu-
line past tense forms for each feminine past tense form. This study explores
verbs at the two extremes of the feminine:masculine scale: the top 100 primarily
masculine verbs and the top 100 primarily feminine verbs. These groups of
verbs provide a concrete basis for assessing cultural stereotypes associated with
gender. This chapter is structured as follows. Section 3.1 introduces the topic of
language and gender. In section 3.2. I discuss the methodology of grammatical
profiling and similar quantitative approaches to language. Section 3.3 desribes
the methodology used in this study. Sections 3.4 and 3.5 present lists of the 100
most masculine and most feminine verbs respectively 4.
Men and women approach the world differently: some views generally pertain
to men and some activities are mostly performed by men, while others are more
preferred by women. This chapter offers a scientific account for this intuition by
introducing a quantitative method of measuring gender stereotypes. This is
possible because a Russian past tense verb form agrees in gender with the sub-
ject. Using the ratio of feminine past tense forms to masculine past tense forms
as an independent measure, we can draw conclusions about the gender pre-
suppositions associated with various activities. Lists of activities undertaken
predominantly by men as opposed to women can be compiled and analyzed.
People tend to see others “as representatives of groups” (Tannen 1990: 15).
One of the most significant grouping parameters is gender, which is “biological
difference paired up with unanalyzed behavioral stereotypes” (Eckert and
McConnell-Ginet 2003: 13). In the last two centuries men and women have be-
||
4 A version of this chapter was published as Kuznetsova 2015; however, this chapter focuses
more on the distribution of gender inflections as an example of the grammatical profiling
approach.
32 | Grammatical profiling and gender stereotypes
ican English. She finds that he has 9,543 occurrences, while she has only 2,859,
which means that men are referred to three times as often as women5.
This ratio is similar to the ratio in the dictionary study by Nilsen (1977). Nil-
sen extracted words with overt sex markers and demonstrated that masculine
words are three times more frequent than feminine words: 385 words with mas-
culine markers as opposed to 132 words with feminine markers. This difference
is similar to the difference in the frequency of the pronouns on ‘he’ and ona
‘she’ in the frequency dictionary of Russian (Zasorina 1977), as reported in
Yokoyama (1986: 154): “on is two and a half times as frequent as ona (on : ona =
13,143 : 5,836).” This misbalance is comparable to the most frequent femi-
nine:masculine ratio found for Russian verbs in section 3.3, where I show that
an average Russian verb has three masculine past tense forms for each feminine
past tense form.
Although there are several dictionary studies that uncover gender misbal-
ance between Russian nouns associated with gender (Martynyk 1991a, 1991b,
Yokoyama 1986, Krongauz 1996), and experimental studies have been conduct-
ed that discuss the gender interpretation of professional nouns (Doleschal 1997,
Schmid 1998), few corpus studies have addressed gender issues, with the excep-
tion of a small corpus study by Kirilina (2002) that investigated collocations of
ženščina ‘woman’ and mužčina ‘man’ in the newspaper Komsomolskaya Pravda
in 1997-1999. The present study is the first large-scale corpus study of verbs that
are strongly associated with masculine or feminine gender in Russian.
||
5 However it is important to keep in mind the fact that these numbers include also gender-
nonspecific generic use of he referring to an indefinite person (contexts like If anyone comes tell
him…).
34 | Grammatical profiling and gender stereotypes
This section describes how the data for this study has been collected, what
items are excluded and the reasons for excluding them. Using the resulting
36 | Grammatical profiling and gender stereotypes
database I calculate the fem:masc ratio and discuss the distribution of the ratios
in the RNC.
For this study I used the database of verb forms from the RNC assembled for
the frequency dictionary of the RNC (Lyashevskaya and Sharoff 2010). This da-
tabase contains all verb forms that occur in the modern subcorpus (after 1950)
of the RNC and their frequencies. In Russian each lexical item is either imperfec-
tive or perfective, and the frequency dictionary treats imperfective and perfec-
tive verbs that have the same lexical meaning as different verbs. Because this
study uses Lyashevskaya and Sharoff’s database as its source, imperfective and
perfective verbs with the same lexical meanings are treated as separate verbs.
I selected all verbs that had past tense forms attested in this database. This
selection gave me 24,872 lexemes. Some of the verbs in this sample have very
low frequency, and thus few examples of feminine and masculine past tense.
Since the relative dominance of one or the other gender form is less meaningful
for verbs of very low frequency, I included in the study only verbs with twenty
or more forms of the singular past tense. The resulting database consisted of
8,340 lexemes (over 6,000,000 past tense forms). All manipulation with the
data discussed below is based on this database.
In order to focus on the gender distribution of verb forms, it was necessary
to remove from consideration impersonal verbs that primarily express neither
feminine nor masculine. Some Russian verbs are used almost exclusively in the
neuter past tense. These are impersonal verbs like rassvetat’ ‘be dawning’ or
podumat’sja ‘occur to smbd’. The question of whether a given verb of this type
has more masculine or more feminine uses is not meaningful for this study. For
all verbs I measured the percentage of neuter uses. For 95 verbs this percentage
was 90% or higher, and these verbs were excluded from the subcorpus for the
study. For the remaining verbs, I calculated the feminine:masculine ratio using
this formula:
where fem (verbX) is the number of occurrences of the feminine past tense end-
ing of a verb and masc (verbX) is the number of occurrences of the masculine
past tense ending of a verb. This gives a ratio ranging from 0 (for verbs with no
feminine past tense uses) to ∞ (for verbs with no masculine past tense uses).
This ratio is rounded to one decimal place. For example, the verb pererubit’
‘chop in two’ has 2 feminine past tense forms and 36 masculine past tense
forms: the fem:masc ratio is 2/36 = 0.05555556 and this number is rounded to
0.1.
Methodology of the study | 37
Once we have a fem:masc ratio for each verb it is possible to examine the distri-
bution of ratios. The highest finite ratio is 90.0 for the verb otvorit’sja ‘open’. In
addition there are 51 verbs with a ratio of ∞ (verbs with no masculine past tense
forms). The lowest available ratio is 0, and there are 46 verbs with this ratio. The
data is skewed towards 0 (most of the verbs have a ratio between 0 and 1 and
few verbs have a ratio more than 1). Thus the most interesting portion of this
distribution is the segment between 0 and 1. This segment is shown in more
detail in Figure 4. The largest number of verbs have a ratio of 0.3, which means
that a typical Russian verb in the RNC has three masculine past tense forms for
each feminine past tense form. This finding coincides with that of Nilsen (1977),
Yokoyama (1986), and Romaine (2001), who observed a similar distribution for
pronouns in both Russian and English. Thus we are dealing with a parameter
that is not necessarily specific to Russian and may be more universal.
Three ratios depicted in Figure 4 are associated with more than 1000 verbs
each, namely ratios 0.2, 0.3 and 0.4. The exact numbers of verbs with these
ratios are presented in Table 7. The sum of the verbs associated with these three
ratios is 3,906.This means that almost half of the verbs explored (3,906 verbs
38 | Grammatical profiling and gender stereotypes
out of 8,340) have a fem:masc ratio between 0.2 and 0.4. The verbs that have a
fem:masc ratio that is lower than 0.2 or higher than 0.4 can be considered
skewed towards masculine or feminine use, respectively.
What is the best way to characterize this data statistically? The mean would
not give a statistically good description of the data for two reasons. First, the
data is skewed towards 0, and the mean is not a good measure of the central
tendency of skewed data. Second, since we have infinity in our data the mean
would also necessarily be driven to infinity since the mean is sensitive to ex-
treme outliers. Therefore the mode and median are better measures. The mode
is the most frequent of the occurring parameters and the median is the parame-
ter that divides the sample into two halves. For this sample, the mode is 0.3 (the
peak in Figure 4) and the median is 0.4 (this means that half of the verbs have a
ratio less than 0.4).
The verb agrees in gender with the subject. Thus it is important to pay attention
to the distribution of overall gender for nouns in the corpus. Table 8 shows how
many nouns of each gender are found in the RNC (using token frequency). The
fem:masc ratio for nouns in the RNC is 0.7. This means that gender distribution
among the verb forms is more skewed than noun distribution predicts. If the
verbal distribution reflected noun distribution, then the mode and median of
the fem:masc ratio for verbs would be 0.7, not 0.3 and 0.4.
Yet, if we limit our search only to nouns referring to humans, the relation be-
tween the feminine nouns and the masculine nouns changes. The fem:masc
ratio for nouns referring to humans is 0.3. Thus the gender distribution of the
past tense form endings reflects the distribution of nouns referring to humans,
which is probably a result of the fact that most frequently the subject of a verb is
human.
Once we know the most typical fem:masc ratio, we can explore verbs at the two
extremes of the fem:masc scale. However, before turning to the verbs that show
strong preference for feminine or masculine past tense use we must identify
other factors that can lead to a very high or a very low fem:masc ratio.
One of these factors is idiomatic usage. An idiom will have one or a limited
set of possible subjects, thus forcing the agreement of the past tense form. This
phenomenon can be illustrated by the verb podrat’ ‘tear up’. Feminine and neu-
ter past tense forms of this verb never occur in the corpus. Out of 364 occurrenc-
es of masculine past tense, 347 are instances of the idiom Čert by ego podral!
‘Damn him!’ (lit. ‘Let the devil tear him up!’), where the subject is the masculine
noun čert ‘devil’.
Another group of verbs with a strong gender preference can be termed
“verbs with a predefined subject.” These are verbs most frequently used with
one particular non-human subject. This phenomenon is frequent within the
class of weather verbs. The verb nakrapyvat’ ‘drizzle’ is a clear example of a verb
with a predefined subject. Its strong affinity for masculine forms is not con-
nected to the meaning of the verb; its gender preference reflects the gender of
its predefined subject which is the masculine noun dožd’ ‘rain’ or its diminu-
tives doždik and doždiček (111 out of 113 occurrences); see Table 10.
Table 10: Distribution of subjects used with feminine and masculine past tense of the verb
nakrapyvat’ ‘drizzle’
The verb otvorit’sja ‘open’, which has the highest non-infinite fem:masc ratio
90.0, is an example of a feminine verb with a predefined subject. The most
frequent subject of this verb is the feminine noun dver’ ‘door’. This verb has 12
attestations of masculine past tense and 1090 attestations of feminine past
tense; see Table 11. All verbs with predefined subjects found in the present
study have non-human dominant subjects and thus are not relevant for this
research. Verbs with a gender distribution that results from idiomatic use or a
predefined subject have been determined manually and are excluded from this
study.
Table 11: Distribution of subjects used with feminine past tense of the verb otvorit’sja ‘open’
The following two sections discuss the top 100 verbs from both ends of the
fem:masc scale. Some of the verbs in the database are compatible with both
human and non-human subjects. In such cases, only examples with human
subjects were analyzed in order to determine whether a particular verb has a
preference towards masculine or feminine use. However, in order to preserve
consistency in the presentation of the data, the ratios for such verbs are report-
ed based on all examples in the modern subcorpus of the RNC.
This section explores the top 100 verbs with human subjects from the masculine
end of the scale, i.e. the top 100 masculine verbs. It is important to note that this
list includes only verbs that collocate with human subjects. For masculine verbs
The top 100 masculine verbs | 41
the top 100 verbs collocating with humans are found in first 175 verbs, meaning
that seventy-five verbs out of the first 175 are idiomatic verbs or verbs with pre-
defined subjects.
The top 100 masculine verbs can be divided into seven semantic classes as
distributed in Table 12. “Role” refers to activities associated with men in society;
“negatively evaluated behavior” refers to verbs of smoking, drinking, and gam-
bling; “talk” denotes a class of verbs referring to participation in a dialogue;
“physical” marks a class of actions that presuppose physical strength; “liturgi-
cal” denotes verbs used in liturgical texts; and the “gesture” class consists of
verbs that name masculine gestures. I will present the verbs that belong to each
of the classes, dividing them into subgroups based on their semantics.
3.4.1. Role
The class “role” can be divided into subgroups of activities, associated with
primarily masculine roles, as in Table 13. This table and all the tables below are
structured as follows: the first column lists verbs, the next column gives their
glosses, the next three columns contain the number of feminine, masculine and
neuter past tense endings, the next column gives the fem:masc ratio for the
verb, and last column contains the subclass of the verb. All verbs that belong to
the same subclass are given in the table together, ordered from the lowest ratio
to the highest (thus the verbs with stronger masculine preference are listed
first). Subclasses are ordered according to the number of verbs they contain
(larger classes are listed first).
There are six verbs associated with leadership. These verbs either mean to
be in charge, like načal’stvovat’ ‘be chief’ or predvoditel’stvovat’ ‘chair’, or pre-
suppose being in a leadership position, like otrjadit’ ‘dispatch, pomilovat’ ‘par-
don’ or kurirovat’ ‘supervise’ – actions that you can perform only if you are in
an executive position. The next five verbs refer to criminal activities of different
42 | Grammatical profiling and gender stereotypes
types. These include illegal activities like robbing, stealing or extorting (ograbit’
‘rob’, umyknut’ ‘go away with’, vymogat’ ‘extort’), as well as activities usually
associated with illlegal behavior such as perelezat’ ‘climb over’, which is usual-
ly used with the object zabor ‘fence’ and refers to the situation of trespassing.
||
6 Smotat’ udočki ‘pull up fishing rods’ has an idiomatic use with the meaning ‘depart in haste’,
however most uses of the masculine past tense refer to the actual handling of fishermen’s
wares.
The top 100 masculine verbs | 43
The next four classes of verbs show that it is men who are usually involved with
horses, fishing, and weapons. Other verbs in the “role” class represent profes-
sions which men perform more often than women, such as conductor (dirižiro-
vat’ ‘conduct’), photographer (zasnjat’ ‘take a snapshot’), priest (otpevat’ ‘per-
form a funeral service’), and soldier (provoevat’ ‘be at war’). Some “role” verbs
refer to activities that men perform more often than women; for example a per-
son who has visited many places in the world is more likely to be a man than a
woman (iz”ezdit’ ‘travel around’). The uses of the verb obrit’ ‘shave off’ are al-
most equally distributed among the contexts associated with shaving the beard
and shaving the head, both usually performed by men.
The verb ssadit’ has two meanings: ‘help down’ and ‘abrade the skin’. Both
of these meanings occur more often with masculine subjects than feminine
subjects, however thirty-two of the thirty-nine uses of the past tense ssadil refer
to ‘helping down’, suggesting that it is the role of a person who helps another
person down that makes the verb ssadit’ a masculine verb; see (5).
The verb naxaživat’ ‘find’ is not a clear case of a dominant masculine use: its
uses are very skewed due to one text in the corpus. Thirty-two of the thirty-six
forms found in the corpus come from a single text – “Zapiski ružejnogo oxot-
nika Orenburgskoj gubernii” (The notes of a hunter from Orenburgskaja prov-
ince) by Aksakov, where the verb refers to the hunter finding wildfowl; see (6).
Because this data is skewed by a single text, this verb is excluded from further
discussion.
44 | Grammatical profiling and gender stereotypes
Fifteen verbs of negatively evaluated behavior can be seen in Table 14; the type
of behavior correlating with a verb is shown in the last column of the table.
(7) Doktor zalpom vypil dve bol’šix rjumki i srazu osovel. [D.N. Mamin-
Sibirjak. Xleb (1895)]
‘The doctor drank two large shots at one gulp and immediately became
drowsy.’
The class includes three verbs of smoking (dokurit’ ‘finish smoking’, raskurit’
‘light up (a cigarette)’ and its imperfective counterpart raskurivat’) and two
verbs of squandering money (prokutit’ ‘dissipate’ and promotat’ ‘waste one’s
money’). It is interesting to note that even though verbs of gambling – another
“bad habit” – do not appear on the Top 100 list, these verbs also have strong
preference for masculine use, consider proigrat’sja ‘lose all one’s money (in
gambling)’:
Table 15: Data for the verb proigrat’sja ‘lose all one's money (in gambling)’
This class presents masculine verbs of talking and argumentation. They can be
divided into four subclasses: verbs that refer to argumentation itself, verbs of
evaluation, verbs that describe manner of speaking, and verbs that designate
sounds usually produced by men; see Table 17.
and two verbs that refer to expounding one’s thoughts (iz”jasnit’ ‘explain’ and
izlagat’ ‘expound’). Verbs of evaluation are closely related to the verbs of argu-
mentation. This subclass contains verbs of comparative evaluation of two ob-
jects (sopostavljat’ ‘compare’), evaluating too highly (pereocenit’ ‘overesti-
mate’), evaluating as bad (raskritikovat’ ‘criticize’), and making an incorrect
judgment during evaluation (nedoponjat’ ‘misunderstand’).
Among the “manner of speaking” verbs we find a verb that means to speak
in jest (balagurit’), a verb pokašljat’ ‘cough’ used in dialog when someone is not
sure what to say (see (8)), and the verb oseč’ ‘cut’ that is most frequently used
for cutting off someone’s speech; see (9).
(8) Xm, xm, - smuščenno pokašljal Služkin. – A ja ved’, Len, tak v tebja
vljublen byl… [Aleksej Ivamov. Geograf globus propil (2002)]
‘Hm, hm – coughed Služkin in embarassment. – And, Lena, I was so in
love with you.’
Two of the verbs of sound denote loud sounds produced mostly by men: proo-
rat’ ‘roar’ and prorokotat’ ‘rumble’. Another two verbs of sound refer to un-
pleasant ways of speaking usually not associated with women: speaking
through the nose (prognusavit’) and speaking in a husky voice (zasipet’), relat-
ing these verbs to the verb of belching in the class of negatively evaluated be-
havior. The last verb in this subclass, krjaknut’ ‘grunt’, can function as a speech
verb, but mostly is used as a marker of pleasure; see (10).
3.4.4. Physical
The “physical” class refers to activities that presuppose physical strength, pre-
sented in Table 18.
48 | Grammatical profiling and gender stereotypes
This class contains five verbs produced from the same root rubit’ ‘cut’ with dif-
ferent prefixes (narubit’, vyrubit’, srubit’, porubit’, pererubit’). These verbs most-
ly refer to cutting wood (11) or knocking down one’s enemies in a fight (12),
which presupposes physical strength. It is important to note that two of the
verbs of this group srubit’ and narubit’ can be used metaphorically to mean ‘get
money’; see (13). However such use is not frequent (for example only five exam-
ples out of 116 uses of srubil ‘cut down’ are of this meaning), so even though
these verbs probably occur in this meaning more often with masculine than
feminine subjects, this meaning does not affect the distribution noticeably.
(11) Nanosil v kadušku, čto v senjax stojala, vody, drov narubil. [Vasilij
Grossman. Žizn’ i sud’ba (1960)]
‘[I] filled the tub that stood in the hall with water and cut wood.’
(12) …Ja srubil ego prijemom sambo. [Jurij Trifonov. Dom na naberežnoj
(1976)]
‘…I knocked him down with a sambo hold.’
This class contains three verbs of hammer use: skolačivat’ ‘hammer smth to-
gether’, vykolotit’ ‘knock out’, zakolačivat’ ‘hammer (nails) in’. All three verbs
are from the root kolotit’ ‘beat’ using different prefixes. There are also verbs
denoting actions involving physical strength such as breaking (vylomat’ ‘break
The top 100 masculine verbs | 49
in’ and razlomat’ ‘break down’), digging (zakapyvat’ ‘dig into smth, bury’), driv-
ing into (vognat’ ‘drive into’), pushing away (rastalkivat’ ‘push away’).
3.4.5. Gesture
This class unites verbs that designate gestures usually performed by men, pre-
sented in Table 19. The last column shows the part of the body that is involved
in the gesture (one verb, naxlobučit’, includes not only a part of the body – the
head – but also a hat).
part of the
lemma gloss fem masc neut ratio body
ponurit’ ‘hang (one's
(golovu) head)’ 1 41 0 0 head
naxlobučit’ ‘pull down (over
(na glaza) one’s eyes)’ 11 152 0 0.1 head
pokručivat’ ‘twist (one's
(usy, moustache, one’s
pal’cami, fingers, one’s mustache,
golovoj) head)’ 1 33 0 0 fingers, head
pobarabanit’ ‘drum (one’s
(pal’cami) fingers)’ 4 58 0 0.1 fingers
potirat’ (ruki)
‘rub (one’s
hands)’ 11 288 1 0 hands
žmurit’ (gla- ‘screw up (one's
za) eyes)’ 1 31 0 0 eyes
pošagat’ ‘start walking’ 6 109 0 0.1 legs
Several masculine gestures reflect the emotion of a person: hanging one’s head
as a sign of depression (ponurit’ ‘hang’), pulling a hat over one’s eyes as a sign
of displeasure (naxlobučit’ ‘pull down’), rubbing one’s hands as a sign of delight
(potirat’ ‘rub’), drumming one’s fingers as a sign of nervousness (pobarabanit’
‘drum’), screwing up one’s eyes as a sign of pleasure (žmurit’ ‘screw up (one’s
eyes)’), and twisting one’s mustache or one’s fingers as a sign of pleasure
(pokručivat’ ‘twist’); see (14).
(14) Vybrav tenistuju berezu, Aniskin upersja v nee plečom, podumav, vynul
ruki iz-za spiny, složil ix na puze i stal pokručivat’ pal’cami s takim
vidom, slovno stojat' pod berezoj bylo dlja nego samoe bol’šoe
50 | Grammatical profiling and gender stereotypes
The verb pošagat’ ‘start walking’, etymologically related to šag ‘step’, has a
tendency to be used more frequently with masculine subjects, because it means
‘to walk sharply, resolutely using wide measured steps’, which is more associ-
ated with men’s behavior than with women’s. Consider the verb prošagat’ ‘walk
a distance’ with the same root which is also skewed towards masculine use: it
has fem:masc ratio 0.1. When women are described with these verbs their be-
havior is more masculine: they are described as fast and resolute (15), angry and
speaking in a deep voice (16), or even locking up a man (17).
Among the eight verbs of this subclass, five are mostly used in liturgical texts.
In such texts the subjects of the sentences are usually god, holy people (includ-
ing monks and priests), or mighty people such as tsars or princes. Therefore the
The top 100 masculine verbs | 51
verbs that are used predominantly in liturgical texts show a preference for mas-
culine past tense.
The verb zapovedat’ ‘enjoin’, for example, is rarely used with the feminine past
tense ending, while it has 109 attestations in the masculine form. There are only
two occurrences in the corpus of feminine past tense uses of zapovedat’ ‘enjoin’:
both have a respected grandmother as their subject and both belong to the 19th
century, so it can be concluded that this usage is rare or non-existent in modern
texts. Almost all uses of masculine past tense refer to god (81 out of 107), be-
cause zapovedat’ ‘enjoin’ is used when one is speaking about what god enjoins
people to do. The verb zapovedat’ is etymologically connected to zapoved’
‘commandment’. God is referred to by several subjects, such as Christ or the
Savior, but all of them have masculine gender, because the concept of god in
Christianity is masculine. Other uses of the masculine past tense of zapovedat’
‘enjoin’ refer to holy people like Moses, or Muhammed, or John the Baptist or
various clerics (17 uses altogether). There are also uses that refer to mighty peo-
ple on earth such as czars and princes (4 uses). There is one reference to the
non-Christian god Buddha, two to otec father (in one of those examples the
father is also a prince), one to xeruvim ‘cherubim (a type of angel)’, and one to
narod ‘nation’ (see example (18)).
(18) Narod zapovedal ne zrja v poslovicax: “Za Rus’ idi vpered, ne trus’ ”.
[Bratstvo krapovyx beretov (2004) // “Soldat udači” 2004.04.07]
‘There is a reason why the nation enjoined [people] in proverbs: “Take
the lead for Russia, don’t be a coward.”’
52 | Grammatical profiling and gender stereotypes
The second subclass of verbs in this group are verbs for which the subject is a
writer, an artist, a sculptor, a composer or another type of creative agent. The
verb obessmertit’ ‘immortalize’ which means ‘use as a prototype for a poem,
picture, sculpture or other cultural object and thus immortalize in the culture’ is
a masculine verb. It is interesting to note that on the rare occasion when the
verb obessmertit’ ‘immortalize’ is used to describe a woman – the writer Dina
Rubina – the masculine noun literator ‘literary man’ is used to refer to her and
the verb has a masculine form, since it refers not to her alone, but to writers in
general:
The verb ottačivat’ literally means ‘sharpen’, but is mostly used in construc-
tions like ottačivat’ masterstvo ‘perfect one’s skill’; see (20).
show us that at least in Russian culture the great majotrity of such artists are
men.
The last verb in this class – the verb derznut’ ‘dare’ is used only in texts of
high style. Compare it with the verb risknut’ which has a similar meaning ‘risk’,
but is less restricted in style. Risknut’ ‘risk’ is more stylistically neutral and has a
fem:masc ratio of 0.2, i.e. is not so strongly skewed towards feminine or mascu-
line use.
Summing up this section on masculine verbs, one third of the masculine verbs
strongly skewed towards masculine use are associated with roles usually car-
ried out by men. Most of these verbs denote the use of professional skills. Thus
we can see that one of the most salient parts of the masculine world seen
through the distribution of verb forms is the professional sphere. These results
are also in agreement with Doleschal’s (1997) and Schmid’s (1998) experimental
studies, which show that hearers interpret nouns that denote professions as
predominantly masculine. If we compare these findings to the top 100 feminine
verbs, which largely lack professional verbs, we perceive a misbalance in the
presumed occupations of men and women. As we will see feminine verbs con-
tain a large subgroup of verbs associated with motherhood and housekeeping,
while no verbs associated with fatherhood and housekeeping appear among the
top 100 masculine verbs.
Aside from professional occupations, the top masculine verbs tell us that
men are physically strong and are able to cut and break objects, and men en-
gage in more negatively evaluated behavior than women. While talking men
use argumentation, which as we will see in section 3.5.6 is absent from the verbs
of feminine talking.
Russian culture recognizes gestures that are specifically produced by men,
with a significant number of gestures signaling pleasure. This tells us that Rus-
sian men show the feeling of delight more openly. We can compare that finding
with the subclass of feminine gestures (see section 3.5.2) which mostly signal
distress. The liturgical verbs and verbs of high style that have god, holy people
and artists as subjects draw our attention to the fact that most spiritual and
artistic areas are dominated by masculine characters. Thus men work, crush
things (using their physical strength), indulge in vices, talk using argumenta-
54 | Grammatical profiling and gender stereotypes
tion, signal delight using their gestures, and dominate in spiritual and artistic
life. Let us now turn to the most feminine verbs and find out what women tend
to do according to the Russian language.
This section describes the top 100 feminine verbs from the fem:masc scale. It is
important to note that this list, like the top 100 masculine list, includes only
verbs that collocate with human subjects. 232 of the first 323 verbs that score
highest on the fem:masc scale collocate with non-human subjects. Thus femi-
nine verbs are much less compact than masculine verbs (cf. for masculine verbs
the first 100 verbs collocating with humans are found in the first 175 verbs). In
other words Russian verbs with primarily feminine past tense forms are more
frequently the result of collocations with one or several primarily feminine inan-
imate nouns.
The top 100 feminine verbs with human reference have ratios from 69.0 to
1.3. The verb with the highest fem:masc ratio is zaberemenet’ ‘get pregnant’
which has the ratio 69.0 and has 207 attestations of feminine past tense vs.
three attestations of masculine past tense, where masculine past tense once is
used for generic reference as in (21), and two other examples are used in a fic-
tion story where a masculine protagonist gets pregnant.
The top primarily feminine verbs can be divided into six semantic classes shown
in Table 23. “Role” refers to activities associated with women in society, includ-
ing maternity, domestic work and witchcraft. “Emotions” denotes a class of
verbs of emotional involvement or emotional response. “Relationship” marks
verbs denoting interpersonal relations. The “appearance” class consists of verbs
that refer to a woman’s appearance. The “movement” class consists of verbs
that name stereotypically feminine movements. Finally, “talk” denotes a class
of verbs that signal participation in a dialogue. Parallel to the presentation of
masculine verbs, I present the verbs of each feminine class and divide them into
subgroups based on their semantics. Verbs within a given subclass are given in
The top 100 feminine verbs | 55
order from the highest to the lowest to reflect the strength of preference for fem-
inine use.
class verbs
role 45
emotions 21
relationships 10
appearance 9
movement 8
talk 7
TOTAL 100
The mostly feminine roles can be divided into two big classes associated with
maternity and housekeeping plus one small class associated with witchcraft.
The first two classes are so large that these three subclasses will each be dis-
cussed separately. The verbs associated with maternity can be further divided
into two subclasses associated with child-rearing and child-bearing.
Seven verbs refer to different crafts, all of which can be characterized as nee-
dlework. Here we can find verbs of sewing (prišivat’ ‘sew on’, prošit’ ‘stitch’ and
obšivat’ ‘make clothes for’) and verbs of embroidery (vyšit’ ‘embroider’ and its
imperfective counterpart vyšivat’) all produced from the same root šit’ ‘sew’, and
verbs štopat’ ‘darn’ and vjazat’ ‘knit’. We can see that while masculine crafts are
associated primarily with woodwork, all feminine crafts are produced by nee-
dles, thus portraying differently men’s and women’s craft production. Two
large subclasses among housekeeping verbs denote cooking and washing.
Cooking is mostly represented by baking (there are peč’ and its two perfectives
ispeč’ and napeč’) accompanied by whisking (vzbivat’ ‘whisk’), a colloquial term
for cooking (strjapat’ ‘cook’) and a verb used for warming up food (razogret’
‘warm up’). Verbs of washing include verbs of washing clothes (stirat’ ‘wash
(clothes)’ and its two perfectives postirat’ and vystirat’) and verbs of washing
dishes (myt’ ‘wash’ and peremyt’ ‘wash all’). There are also two verbs of ironing
(vygladit’ and nautjužit’), two verbs of milking (podoit’ and its imperfective
counterpart doit’ ) as well as of cleaning (pribirat’), whitewashing (pobelit’),
setting a table (nakryvat’ (na stol)), weeding (polot’) and the verbs stelit’ and
postelit’ ‘lay’ which can refer to two different household activities, which both
occur in the corpus with the feminine past tense (po)stelila: (po)stelit’ skatert’
means ‘lay a tablecloth’ and (po)stelit’ postel’ means ‘make a bed’. Thus we see
that twenty-six verbs of housekeeping appear in the top 100 feminine verbs,
while none appear in the top 100 masculine verbs. This shows that according to
the Russian cultural prototype a woman’s role includes a lot of domestic work
as can be portrayed by example (22), where a woman combines work with de-
manding domestic and medical duties.
(22) Marina plastalas’ posle raboty i v vyxodnye dni, pobelila vse, čto nado,
tože učas’ v processe žizni i po podskazkam xozjajki, perestirala, pere-
trjasla vse baraxlo v dome i medicinskie objazannosti spravljala
snorovisto i umelo. [Viktor Astaf’ev. Proletnyj gus’ (2000)]
‘Marina worked hard after work and on the weekends, whitewashed
everything that needed it, learning in the process and using tips from
the landlady, she washed everything, rummaged through things in the
house and performed her medical duties nimbly and skillfully.’
58 | Grammatical profiling and gender stereotypes
The subclass of withcraft verbs contains only two verbs: nagadat’ ‘tell fortunes’
and privorožit’ ‘bewitch’. The appearance of these two verbs among the top 100
feminine verbs highlights a cultural stereotype according to which women are
connected with magic and superstition, where a woman can predict one’s future
or even bewitch a man using her charms.
It is important to notice that while the masculine role class mostly consists of
verbs denoting various professions, women’s roles only portray women as
mothers, housewives or witches. These findings support Smith (1985: 56) who
concludes that, as opposed to men who are believed to be active in the public
sphere, “[w]omen are encouraged to be responsible for the integrity of the do-
mestic sphere of life, including working at home and caring for children.” The
first feminine verb referring to an actual profession that appears on the
fem:masc scale is the verb perepečatyvat’ ‘retype’ and denotes the profession of
a typist as typically feminine in Russia, which is also reflected in the name of
the profession mašinistka7 ‘typist’ which includes feminine syffix -ka. However
this verb ranks as number 101 among the verbs referring to people from the
feminine end of the scale, and therefore it is not included in the top 100 list.
Therefore we can see a clear distribution of social roles determined by gender:
while the prototypical man spends his time at work, the prototypical woman
spends her time at home caring for children and doing domestic chores.
3.5.2. Emotions
This section describes verbs associated with emotions. These verbs are primarily
feminine, and note that the top 100 masculine list did not include any verbs
referring to similar emotions.
||
7 Even a man employed as typist is referred to as mašinistka. Usually there are both masculine
and feminine nouns referring to a profession that denote the same occupation, as in svjazist
‘male postal and telecommunications worker’ and svjazistka ‘female postal and telecommuni-
cations worker’. However the masculine noun mašinist, though it exists, is not connected to
typing and means instead ‘train engineer’.
The top 100 feminine verbs | 59
Tears are the most frequent sign of an emotion expressed by women. This can
be seen from the most dominant subclass in this class that consists of nine verbs
of crying. They include crying (vsplaknut’ ‘have a little cry’, proplakat’ ‘cry for a
certain time’, vyplakat’ ‘cry out’, poplakat’ ‘cry for a while’), weeping (prorydat’
‘weep something’, zarydat’ ‘start weeping’), sniffling (zaxljupat’ ‘start sniffling’,
zašmygat’ ‘start snivelling’) and sobbing (vsxlipnut’ ‘sob once’). Closely related
to the verbs of crying are the verbs of lamentation. Women can lament (pričitat’
and zapričitat’) or wail (zagolosit’ and its imperfective counterpart golosit’).
Another sign of emotion that women frequently use is exclamation. They can
exclaim producing various vowel sounds such as ah! (zaaxat’ ‘start exclaim-
ing’), oh! (pooxat’ ‘sigh oh! for a while’), and oy! (ojknut’ ‘say oy’) or a squeal
(vizgnut’ ‘squeal once’).
60 | Grammatical profiling and gender stereotypes
Women also show emotions by gestures. Note that while masculine ges-
tures often signal greed or contentment, feminine gestures reflect emotional
distress. There are two main strategies of showing emotions by gestures found
in the top 100 list: one is lifting hands as a sign of deep distress (vsplesnut’
(rukami) ‘lift (one’s hands)’ and its imperfective counterpart vspleskivat’) and
another is becoming completely numb and not showing any emotions at all
(zakamenet’ ‘be petrified’, omertvet’ ‘become immobilized’); see (23)-(24).
(23) Ona kak budto omertvela, obledenela, vse vokrug kazalis’ ej ne-
dobroželateljami. [Dar’ja Simonova. Sorvannaja sliva (2002)]
‘It was as if she went numb and icy, everybody seemed to be against
her.’
The last option for showing emotions available to women is cheering up. There
exists a specifically feminine gesture by which a woman can try to cheer up her
interlocutors by gently shaking them; see (25)-(26).
Thus we can see that in the sphere of stereotypes only women show emotions of
distress. Crying, exclaiming and lamentation are considered to be feminine
behavior that is rarely shown by men.
3.5.3. Relationships
Verbs associated with relationships contain five verbs of delight, four verbs of
negatively evaluated behavior, and one verb of positively evaluated behavior.
The top 100 feminine verbs | 61
Most verbs in this class describe aspects of relationships between a man and a
woman.
The biggest subclass of verbs of relations are those relating to delight. Three
verbs in this class (umilit’ ‘delight’, its imperfective counterpart umiljat’ and
vosxiščat’ ‘enrapture’) describe a woman as the stimulus of delight experienced
by a man. Two other verbs characterize a woman’s positive response to such
delight. In reacting to a man’s delight a woman can actively flirt (koketničat’) or
passively cast down her eyes (potupljat’ (vzor)). It is interesting that verbs de-
scribing positive male/female relationships occur in the top list only for verbs
with feminine subjects, i.e. those verbs that have a woman as an agent or a
stimulus. The top 100 masculine verbs contain several verbs describing rela-
tionships between a man and a woman, but those are verbs of sexual aggres-
sion; see 3.4.2.
Four verbs present a subclass describing negatively evaluated feminine ac-
tions. Like the subclass of “delight”, these verbs are often used to describe fem-
inine/masculine relationships. The verb otšit’ ‘rebuff’ most frequently refers to a
woman who rebuffs a man making a pass at her. Razlučit’ portrays a woman
whose behavior affects other people’s relationships: she is separating them,
usually out of jealousy. The verb bludit’ ‘play whore’ is used for a woman who is
not selective in her relationships with men, which is negatively evaluated; see
(27). It is interesting that another verb in the relationships class describes the
opposite behavior: bljusti (sebja) ‘guard (oneself)’ means that a woman shows
appropriate behavior in her relationships with men, as in example (28).
62 | Grammatical profiling and gender stereotypes
(28) Zojka bljula sebja, stalo byt’, pomnogu na glazax načal’stva ne zapivala,
s ženatymi mužikami ne guljala. [Viktor Astafiev. Pečal’nyj detektiv
(1982-1985)]
‘Zojka behaved decently, so she did not drink much when the bosses
were looking and did not go out with married men.’
As noted above, the top 100 masculine verbs contain several verbs of aggressive
behavior. Thus in such situations a man plays the role of an aggressor. Among
the feminine verbs there is one that describes how a woman can solve a conflict.
Since she is weaker, she can use the judiciary system and get something
through a court order (otsudit’), as in example (29).
3.5.4. Appearance
Nine of the 100 top feminine verbs are associated with feminine appearance.
Three verbs refer to changes in appearance. One verb denotes that a person
looks worse than before (podurnet’ ‘look worse’), and two verbs (poxorošet’
‘look better’ and its imperfective counterpart xorošet’) point out that a person
looks better than before. Two verbs produced from the same root, krasit’ ‘paint’,
refer to application of make-up: nakrasit’ ‘make up’ and podkrasit’ ‘touch up’.
Two verbs refer to braiding of hair: zaplesti ‘braid’ and its imperfective counter-
part zapletat’. Two verbs refer to items of feminine apparel: the verb opravljat’
‘adjust’ refers to a dress, and the verb zacokat’ ‘start clicking’ refers to the click-
ing sound made by high-heeled shoes; see (30).
3.5.5. Movement
The class of movement verbs contains five verbs with the root porx- ‘flit’ (porxat’
‘flit’, vporxnut’ ‘flit away’, porxnut’ ‘flit once’, uporxnut’ ‘flit away’, vyporxnut’
‘flit out’). Thus Russians compare women’s movement in space with the move-
ment of birds. The concept underlying this comparison can be formulated as
follows: women are small and light as birds and therefore they move like birds
and their movement can be described as flitting. WOMEN ARE BIRDS functions as a
64 | Grammatical profiling and gender stereotypes
The class of movement verbs contains also two verbs of floating (vplyt’ ‘float
into’ and its imperfective counterpart vplyvat’). These uses compare the move-
ment of a woman in a long dress with floating. Rakhilina (2007: 275) in her
study of Russian swimming and floating verbs notes that verbs of floating in
Russian can be used metaphorically if the movement is similar to floating, i.e. it
is “a slow calm unnoticeable movement, where the object is merging with the
background surface as if the movement is caused by the surface itself” 8. The
slow movement of a woman in a long dress satisfies this condition since the
observer cannot see the woman’s feet, and therefore the woman appears to the
observer to be floating on the surface of the floor; see (31).
||
8 “[Н]ебыстр[ое] спокойн[ое] движени[е], незаметн[ое] глазу, при котором движущийся
предмет сливается с поверхностью-опорой так, как будто движение происходит за счет
самой опорной поверхности” – translated from Russian by J.K.
The top 100 feminine verbs | 65
The verb prošuršat’ ‘rustle past’ is derived from the root šuršat’ ‘rustle’ which
refers to the sound that a long dress with many petticoats makes when a woman
is walking; see (32).
We can see that women’s movement can be divided into two subgroups: verbs
that describe woman’s movement as flitting using the metaphor WOMEN ARE BIRDS
and verbs that specifically describe women wearing long dresses – in which
case a woman can be described as floating or rustling. Note that the only verb
that could be classified as a verb of movement among the top 100 masculine
verbs is the verb pošagat’ ‘start walking’ which portrays walking using wide
measured steps. This shows that Russian has clear-cut notions about both mas-
culine and feminine movement, where masculine movement is sharp and delib-
erate, while feminine movement is light and sporadic.
3.5.6. Talk
The seven verbs that characterize women’s speech all compare women with
various birds. While women are the target domain in this series of metaphors,
the source domains are different kinds of birds. Ščebetat’ ‘twitter’ and zaščebe-
tat’ ‘start to twitter’ refer to small birds, kudaxtat’ ‘cluck’ and zakudaxtat’ ‘start
to cluck’ compare the subject to a hen, taratorit’ ‘chatter’ and zataratorit’ ‘start
to chatter’ characterize the sound produced by magpies, provorkovat’ ‘coo
something’ refers to a pigeon.
When used to refer to people ‘twitter’ (ščebetat’ and zaščebetat’) describes the
cute speech of a young girl. Clucking (kudaxtat’ and zakudaxtat’) describes the
speech of an elderly, possibly plump matron, who cares about her addressees as
a hen cares about its chicks; see (33).
‘Chatter’ (taratorit’ and zataratorit’) describes rapid and not very articulate
speech. These verbs compare women's speech with the noise produced by mag-
pies, wich is importunate and annoying. Cooing (provorkovat’) refers to the
affectionate speech used by lovers.
Thus we can see that Russian characterizes women’s speech through the
sounds usually produced by birds. In other words the metaphor WOMEN ARE BIRDS
that was discussed concerning verbs of movement also plays an important role
in the domain of talking. The verbs of talking reveal traditional men’s stereo-
types about women according to which women’s talk is seen as meaningless or
even importunate and annoying. This supports a common belief that women
talk too much, as illustrated by Russian proverbs. The Dal’ (2009) collection of
ninetineth century Russian proverbs contains the following examples referring
to women’s speech:
These Russian proverbs have analogues in many languages, for instance, they
Summary | 67
can be compared with English Foxes are all tail and women are all tongue, which
shows that this stereotype is shared by other cultures9.
Summing up this section on feminine verbs, the top 100 list of feminine
verbs shows that according to stereotypes a woman’s role does not include a
profession, instead a woman is seen as a mother and housewife, and may also
practice witchcraft. Furthermore the feminine list, unlike the masculine list,
includes verbs of emotions, specifically feminine movements, relationships and
appearance. So we can conclude that it is women who express emotions, wom-
en who care about their appearance and women who are more interested in
participating in a relationship, usually with men. A woman moves and speaks
differently and is perceived as a bird, because she is moving as lightly as a bird
and because what she is saying may sound beautiful, but may also be incom-
prehensible and annoying. Both verbs of movement and speech verbs show us
that Russian uses a conventional metaphor WOMEN ARE BIRDS.
3.6. Summary
||
9 It is interesting to note that according to the findings of Eakins and Eakins (1978), in public
speaking men talk more often than women, and even for a longer time than women do. On the
contrary in private speech (on the phone, in social situations, with friends) it is women who
talk more than men. So men think that women talk more “only because they hear women
talking in the situations where men would not” (Tannen 1990: 78). Thus this stereotype is not
supported by empirical evidence.
68 | Grammatical profiling and gender stereotypes
Some verbs that appear in the top lists were unexpected, leading to unantic-
ipated discoveries. For example it was found that pošagat’ ‘start walking’ is a
verb with strong preference for masculine subjects. This preference becomes
more understandable if we take into account the fact that this verb is associated
with walking sharply, resolutely using wide measured steps which favors mas-
culine subjects. Among the top 100 list of feminine verbs there are two verbs
connected with witchcraft, which reflects the stereotype of women’s association
with the supernatural world. Thus, the picture shown to us by the distribution
of the verbal inflections is familiar in some ways, while unexpected and reveal-
ing in others.
While this picture tells us about common stereotypes, it does not reflect the
actual distribution of roles in the society. In reality Russian women are certain-
ly involved in professional activities, while men participate in the raising of
children. For example the verb rabotat’ ‘work’ has a ratio 0.4 (remember that
0.2, 0.3 and 0.4 are the three most common fem:masc ratios for Russian verbs).
The fem:masc ratio for the phrase V s det’mi ‘V with children’, where V stands
for any verb, is also 0.4. This shows that the amount of work and the number of
activities that women and men share with children are distributed like average
verbs.
This study opens several directions for further research. First, it highlights
the need for more specific research on classes of verbs that show both feminine
and masculine preference. Psychologists studying human gesture have long
known that gestures have strong gender preferences (see, for example Rekers,
Amaro-Plotkin and Low 1977, as well as more recent articles by Briton and Hall
(1995) and Özçalışkan and Goldin-Meadow (2010)). Now such studies can find
support from linguistics. Second, similar studies of other parts of speech, which
contain an overt gender marker, are possible. For example, Russian adjectives
agree in gender with the noun and may give us another perspective on gender
stereotypes. Third, it is possible to explore other languages with verb forms that
mark gender. This could include languages closely related to Russian, such as
for example Byelorussian, as well as unrelated languages such as Hebrew.
Comparing different languages will facilitate a cross-cultural comparison of
gender-related stereotypes.
4. Semantic profiling, predictability and
prototypicality
The previous chapter explored how the meaning of a word is correlated with the
distribution of its inflectional endings. In this chapter a similar approach is
applied to the distribution of a derivational affix and its submeanings. This
chapter examines the polysemous Russian prefix pri- and investigates how the
radial category of the prefix pri- is structured. The radial category is established
in section 4.1. I use semantic profiling and a computational approach to meas-
ure the predictability of the submeanings of the prefix pri- given the meaning of
the base verbs (section 4.2), and then attempt to establish the prototype of the
prefix pri- using features of the prototype (section 4.3). Section 4.3 shows that
different characteristics of the prototype point to different submeanings of the
prefix pri-. Modern cognitive linguistics faces “the necessity to investigate the
phenomenon of prototypicality more thoroughly,” since “the notion is far from
straightforward” (Gilquin 2006: 180-181) and the prefix pri- presents a case of
inconsistency among characteristics of the prototype.
Polysemy is natural for language. Every linguistic unit that has a meaning is
polysemous to some extent. Langacker (2008: 225-226) describes what has to be
learned by a speaker in order to master the meaning of a word:
In learning to use the word properly, a speaker masters the entire network (not just the
schema or the prototype). A lexical item of any frequency tends to be polysemous, having
multiple senses linked by relationships of categorization. Its various senses are members
of a category that is structured by these relationships. It is further said to be a complex
category because its membership and configuration are not reducible to (or predictable
from) any single element.
and periphery. Lakoff (1987: 379) formulates the following characteristic princi-
ples of a radial category:
b. The noncentral subcategories are motivated (italized by Lakoff) by the central mem-
ber; they are neither predictable nor arbitrary.
Lakoff (1987: 83) shows how the radial category functions using the famous
example of the concept mother. The prototype of a mother includes three as-
pects: biological, psychological and social. However the term mother can be
used even if only one of those conditions is true, and the two others are false:
This phenomenon is beyond scope of the classical theory. The concept mother is not clear-
ly defined, once and for all, in terms of common necessary and sufficient conditions.
There need be no necessary-and-sufficient conditions for motherhood shared by normal
biological mothers, donor mothers (who donate an egg), surrogate mothers (who bear the
child, but may not have donated the egg), adoptive mothers, unwed mothers who give
their children up for adoption, and stepmothers. They are all mothers by virtue of their re-
lation to the ideal case, where the base models converge. That ideal case is one of the
many kinds of cases that give rise to prototype effects.
The submeanings of pri- can be divided into four major subclasses (Dobrushina,
Melina, Paillard 2001, Jakunina 2001, Endresen et al. 2012); see Table 32.
The meaning ARRIVE is realized in combination with motion verbs and transpor-
tation verbs. The core of the meaning ARRIVE can be illustrated by the examples:
prijti ‘come’ from idti ‘go’, priletet’ ‘fly to’ from letet’ ‘fly’, primčat’sja ‘come
tearing along’ from mčatsja ‘race.’ This subclass also contains verbs that denote
the final stage of the process expressed by their imperfective counterparts. Here
the idea of ARRIVE is realized within the event structure: these verbs mean that
the result of the process is approached, as in priravnjat’ ‘equate’ from ravnjat’
‘compare’.
The next important subclass for the prefix pri- is ATTACH. These verbs mean
that a trajector becomes closely connected to the landmark. This connection
can be unbreakable, because the landmark is glued (prikleit’sja, prilepit’sja),
chained (prikovat’), nailed (pribit’, prigvozdit’) or by other means is tightly at-
tached to the landmark. The connection can also be temporal and easily break-
able. Here we see verbs of leaning, such as privalit’sja ‘lean’ (from valit’sja ‘fall’)
and pristavit’ ‘put against’ (from stavit’ ‘put’), verbs of breakable contact such
as privjazat’sja ‘be attached’ (from vjazat’sja ‘be tied’), pričalit’ ‘moor to’ (from
72 | Semantic profiling, predictability and prototypicality
čalit’ ‘moor’), and verbs of cuddling including such verbs as prižat’sja ‘cuddle’
(from žat’sja ‘huddle’), priniknut’ ‘press oneself close’ (from niknut’ ‘droop’),
and prigret’ ‘give shelter’ (from gret’ ‘warm’).
This ATTACH subclass also contains a small subgroup of verbs of smoothing
which is best exemplified by the perfective prigladit’ ‘smooth’ (produced from
gladit’ ‘stroke’), which means that the object becomes more smooth and orderly.
The same meaning of the prefix is realized in the verbs primjat’ ‘flatten’ (from
mjat’ ‘rumple’), and pričesat’(sja) ‘comb oneself’ (from česat’(sja) ‘comb one's
hair’). These verbs are grouped together with verbs of attachment because
smoothing and flattening usually involve attachment of bristling parts to the
whole. The verb pribrat’ ‘tidy up’ (from brat’ ‘take’) realizes the more abstract
meaning of making something more orderly.
All verbs of the subclass ADD share the meaning of adding a smaller amount
to a larger group. The semantics of ADD can be clearly seen in such uses of pri-
as prirasti ‘increase’ from rasti ‘grow’ where the object increases by growing,
priselit’ ‘squat’ from selit’ ‘settle’ where new settlers are added to those already
present, pripisat’ ‘add to writing’ from pisat’ ‘write’ where something is added
to what is already written.
The semantic subclass ATTENUATE is characterized by the meaning ‘attenuate
an action’. The result anticipated by the action is not achieved, however some
steps towards it are completed. This meaning can be illustrated by such verbs as
priglušit’ ‘muffle’ produced from glušit’ ‘deafen’ and pritormozit’ ‘slow down’
produced from tormozit’ ‘brake’. Here we see that a perfective verb describes a
movement towards the result of the action, however the result is achieved only
partly: not complete deafening, but just muffling, not a complete stop, but just
slowing down.
The meaning ATTENUATE is closely related to the meaning ADD. The meaning
ADD signals that a small amount of object is added to the larger group of the
same object. Here we see a parallel situation with the quanta of an action. Verbs
with the prefix pri- here denote that small quanta of the activity are added,
however this amount is not large enough to achieve the result. The verbs of the
subclass ATTENUATE may be described as Complex Act perfectives (in terms of
Janda 2007), since here the prefix pri- presents not a change in meaning, but
rather a change in Aktionsart. Figure 6 presents the radial category of the prefix
pri-.
Radial category of Russian prefix pri- | 73
The submeanings in the radial category are connected as follows. The submean-
ing ARRIVE is connected to the submeaning ATTACH, because attachment can be
seen as final point of arrival, where the trajector comes into a contact with the
landmark. The submeaning ATTACH is related to the submeaning ADD, since add-
ing can be seen as attachment of a smaller amount of substance to a larger
amount of substance. The submeaning ATTENUATE refers to adding a smaller
amount of an event to the larger amount of an event; this connects it to the
submeaning ADD which describes adding a smaller amount of substance to a
larger amount of substance. Finally the submeaning ATTENUATE refers to accu-
mulating smaller amounts of an event in order to reach the result, while the
submeaning ARRIVE can be used to describe reaching the resulting stage, which
connects these two submeanings.
There are two main questions concerning the internal structure of this radial
category that are discussed in this chapter. First, I explore how predictable the
submeanings are given the meaning of the base verb, or in other words to what
extent the prefixed verbs with pri- are compositional. Each base verb when
attached to a prefix focuses on a submeaning of a prefix. This raises the ques-
tion of whether 1) the submeaning of a prefix is chosen based on the semantics
of the base verb, or 2) a prefixed verb functions as a “prefabricated unit” and
the submeaning of the prefix is specified for a given prefixed verb. In this study
I use a computational approach to show that neither of these two hypotheses is
completely true. Section 4.2 shows that for approximately half of the verbs with
the prefix pri- the submeaning can be successfully predicted given the meaning
of the base verb.
More light on the internal structure of the radial category of the prefix pri- is
shed in section 4.3, which poses the question: What is the prototype of this
radial category? It is shown that different characteristics of the prototype indi-
74 | Semantic profiling, predictability and prototypicality
cate different submeanings of the prefix pri-. Moreover the distribution of the
submeanings for highly frequent and less frequent verbs with the prefix pri-
indicates that verbs with the submeaning ARRIVE are more likely to function as
prefabricated units, while verbs with the submeaning ATTENUATE are more likely
to be derived.
When investigating the interaction of morphemes within a word, one can ask
whether the meaning of a word is formed compositionally from the meanings of
its morphemes. Within cognitive linguistics this question is better formulated as
“to what extent the meaning of a word is formed compositionally from the
meanings of its morphemes,” since compositionality is seen as a scale, not as a
dichotomy, cf. (Langacker 2008: 245):
In order to understand how the submeaning of the prefix interacts with a verb, I
have conducted a pilot experiment where the speakers were asked to ascribe
meaning to contexts where the verb was omitted. For this experiment I used
four verbs with the prefix pri- that are attested in the RNC in one example each
(according to the frequency dictionary based on the RNC (Lyashevskaya and
Sharov 2010)). These are verbs prišipit’sja ‘huddle’, prifoloxat’sja ‘drag oneself
to’, prišurenit’ ‘screw one’s eyes’ and prišilit’sja ‘be held up’. The fact that there
is only one attestation of each verb in the corpus reflects that these verbs have
low frequency in Russian and are most likely not familiar to the most speakers
of Russian. Compare this number to the raw frequency of the motion verb with
the prefix pri- prijti ‘come’, which is attested in 112,102 examples, or the verb
prikleit’ ‘glue’ with a more specific meaning, which is found in 836 examples.
The first question of the pilot experiment is designed to check whether the in-
formants are familiar with the four verbs. Thirty-four adult native speakers of
Russian have answered the questionnaire of this pilot study. Out of those thirty-
four people, seven reported that they are familiar with the verb prišipit’sja, two
Predictability of the submeaning | 75
reported that they are familiar with the verb prišilit’sja, one did not answer
whether he or she is familiar with the verb prifoloxat’sja, while all thirty-three
people who answered this question are not familiar with this verb and no one in
the study is familiar with the verb prišurenit’. The information on familiarity
with the four verbs is summarized in Figure 7. The columns show number of
informants who are familiar and unfamiliar with each verb.
The informants were given the short description of the four submeanings of the
prefix pri- with several examples for each submeaning. The four examples con-
taining the four verbs mentioned above are culled from the corpus. The verb in
the example was omitted and replaced with the pattern pri____ENDING indicat-
ing that the verb contains the prefix pri- and showing the informant the verb
ending. The verb ending was included so that the informant would be able to
match the verb with its subject and as a result would understand the context
better. The informants were presented with the examples shown below as con-
texts 1-4 and were asked to guess what submeaning of the prefix pri- is used in
this sentence. Information in the gloss of the omitted verb shows the meaning
of the verb ending in the examples.
Figure 8 shows the distribution of the answers for Context 1. Context 1 originally
contained the verb prišipit’sja, the verb that was familiar to seven out of the
thirty-four respondents. The respondents who are familiar and who are not
familiar with the verb are shown separately in this figure.
Context 1 has the most diverse responses in among the four verbs in the experi-
ments. However, from this diverse distribution, we can see that familiarity with
the original verb does not affect the choice of the prefix. Figure 8 demonstrates
that the answers of those informants that were familiar with the verb are dis-
tributed similarly to the answers of those who did not know the original verb
(the difference between the two groups is not statistically significant, Fisher test
shows P = 0.7634.). This means that the respondents did not try to use their
knowledge of the verb to deduce the correct submeaning of the prefix, but on
the contrary, all informants were trying to ascribe the submeaning of the prefix
Predictability of the submeaning | 77
based only on the context. Therefore in Figure 9 respondents familiar and not
familiar with the original verb are united and shown together.
Figure 9 demonstrates that native speakers have clear preferences for one of the
submeanings in all four contexts. Even for context 1, where the informants
chose mostly diverse answers, 47% of them matched the context with the sub-
meaning ATTACH. This is the submeaning that would be best compatible with the
verb prišipit’sja ‘huddle’ originally used in context 1. The remaining three con-
texts show even more clear preference for a submeaning. 91% of all informants
matched context 2 with the submeaning ARRIVE most natural for the verb pri-
foloxat’sja ‘drag oneself to’ originally used in the context 2. 85% of informants
assigned submeaning ATTENUATE to context 3. This is the most likely submean-
ing for the verb prišurenit’ ‘screw one’s eyes’ originally used in that context.
74% of all informants chose the submeaning ATTACH for context 4. This sub-
meaning would be the most natural for the verb prišilit’sja ‘be held up’ original-
ly used in context 4.
Thus, summing up the results of the pilot study, we can say that native
speakers of Russian are able to use context to predict the submeaning of the
78 | Semantic profiling, predictability and prototypicality
prefix even if 1) the verb in the sentence is not shown to them, and 2) they do
not know the verb originally used in that context. Thus, we can conclude that in
the minds of the speakers the context is related to the submeaning of the prefix.
This raises the question of how native speakers of Russian are able to pre-
dict the submeaning of the prefix based on the context. I believe that native
speakers use context to determine a frame employed in the context (cf. Fillmore
and Atkins 1992), and are able to use the frame to predict the semantic class of a
verb. Ultimately, using combined information on the frame and the semantic
class of a verb, native speakers can predict the submeaning of the prefix. The
native speakers are able to use the form of the given sentence to predict the
submeaning of the prefix, so we are dealing with a relationship between form
and meaning existing in speakers’ minds. Unfortunately, frame-tagged data for
Russian prefixed verbs is not available, so a computational model for deducing
the semantic class from the frame is not possible at this point. In this chapter, I
model only the second part of this process – deducing the submeaning of the
prefix based on the semantic class of the verb.
In this chapter I measure the degree of divergence from compositionality we
observe among the pri-verbs and explore what place the verbs with the prefix
pri- occupy on the scale of compositionality using a computational approach.
Computational approaches to language have recently become popular in cogni-
tive linguistics, see for example works on language games by Luc Steels and his
group (Steels 2011, 2012). A computational approach allows us to test proposed
linguistic hypotheses. Even though the success of the computational model
does not allow us to make any conclusions as to whether the human mind func-
tions like the proposed computational model, it is possible that the human
mind uses the same parameters in order to produce a grammatically correct
output.
In this regard the computational approach can be compared to the famous
Kon-Tiki expedition of Thor Heyerdahl (described thoroughly in Andersson
2010). His goal was to prove the possibility of contact between South Africa and
Polynesia – an idea that was inspired by native legends and archeological evi-
dence. In order to prove that it was possible for Inca rafts to reach Polynesia,
Thor Heyerdahl and five fellow travelers constructed a raft, crossed the Pacific
Ocean and successfully reached the Tuamotu islands. This trip proved that a
primitive raft is able to sail the Pacific and cover tremendous distances by fol-
lowing the prevailing wind. However even though the possibility of such con-
tact is proven, Heyerdahl’s expedition does not prove that people from South
America did in fact settle in Polynesia.
Neither does the success of a computational model tell us whether the pa-
rameters that according to our hypothesis govern the distribution are those that
Predictability of the submeaning | 79
enable a speaker of a language to use a linguistic item correctly. The model only
shows that use of the proposed parameters mimics the function of a speaker of a
language and can measure the extent to which the model makes correct predic-
tions. This chapter proposes a computational model for selection among the
submeanings of the Russian prefix pri- and shows how well such a model pre-
dicts the actual behavior of the pri-verbs.
Let us assume that we have a special pri-robot, and that given the base verb and
its semantic tag the robot may apply some rules prescribed for it, such as “If I
see a verb of impact, I choose the submeaning ATTACH” or “If I see a speech verb
which is also non-causative, I choose the submeaning ADD.” In this chapter I
discuss what rules this robot needs to predict the correct submeaning as often
as possible using only the data on the semantic tags and morphological struc-
ture, and what amount of the data can be predicted by the robot. If for most of
the verbs the submeaning of the prefix can be predicted, we can conclude that
the submeanings of the prefix result from combination of the meanings of the
base verbs with the semantic schema of the prefix.
It has to be noted that, even though for a robot it is more natural to function
in terms of “if-then” rules, the human mind most likely does not work this way.
Cognitive research shows that organization of the human brain is best described
in terms of schemas, see for example work by Gibbs and Colston (2006: 260)
who provide “a small part of experimental data that might be related to how
image schemas and their transformations mediate and constitute different as-
pects of cognitive functioning.” However, for the purposes of this chapter the
difference between these two representations is not significant, so the pri-robot
presented in this chapter is using “if-then” rules.
This computational model uses a semantic profiling method, described in
Janda and Lyashevskaya (2013). This method estimates how well semantic tags
of the bases predict the distribution of competing affixes. The correlation be-
tween semantic tags of the bases and types of affixes may be used as another
argument for “meaningful grammar.” Here the distribution of affixes, which
should belong purely to the domain of morphology, is driven by meaning.
For this study I compiled a database of all the verbs compatible with the
prefix pri- listed in the frequency dictionary of the modern subpart of the RNC
by Lyashevskaya and Sharov (2010). This database is divided into two subparts
that are investigated separately in the two following sections. The division is
based on the raw frequency of the investigated pri-verbs. The first subpart con-
80 | Semantic profiling, predictability and prototypicality
tains the verbs that have a frequency of 100 or more in the modern subcorpus of
the RNC, while the second part contains the verbs that have a frequency of less
than 100. A frequency of 100 is used as a threshold between more representa-
tive and more marginal verbs of Russian, cf. the same frequency of 100 in the
modern subcorpus of the RNC is used to choose more representatives verbs of
Russian in the studies by Baydimirova 2010, Janda and Nesset 2010 and Nesset,
Janda and Endresen 2011. Verbs with a frequency of 100 and more from now on
are referred to as “frequent verbs,” whereas the verbs with a frequency of less
than 100 are referred to as “infrequent verbs.”
The robot uses the semantic tags marked in the RNC and derivational tags,
which inform the robot whether a base verb is prefixed or not. The information
on the morphological structure (prefixed vs. non-prefixed base) was produced
automatically and then checked manually. Table 33 shows the semantic and
derivational tags that have at least six verbs among the base verbs that are
compatible with the prefix pri-. The remaining semantic tags represented
among the base verbs were associated with only one or two verbs each and are
not expected to significantly affect the distribution of the submeanings.
This section describes what rules the pri-robot needs to mostly successfully
predict the submeanings of the frequent pri-verbs. It is shown that only one of
the two derivational tags predicts the choice of the submeaning, while five of
the six semantic tags predict the submeaning of the prefix pri-.
I start by investigating the distribution of the submeanings among two der-
ivational tags: the tag “pref”, marking that the verb is produced from a verbal
stem containing a prefix (e.g. otkryt’ ‘open’), and the tag “nonpref”, marking
that the base verb is a simplex verb (e.g. varit’ ‘cook’). Only the tag “pref” signif-
Predictability of the submeaning | 81
icantly affects the distribution of the submeanings, while the tag “nonpref” has
the same distribution as the pri-verbs overall.
Table 34 shows the distribution of the submeanings for the verbs derived
from the prefixed and non-prefixed bases. The second and third columns of
Table 34 contain the actual number of verbs of each submeaning for the verbs
derived from a prefixed base (second column) and a non-prefixed base (third
column). The last two columns of Table 34 contain the expected distribution of
the same numbers. The columns “expectation” show how the numbers would
be distributed if the morphological structure of the base verb did not affect the
distribution of the submeanings. Altogether the database of the frequent pri-
verbs contains thirteen verbs derived from a prefixed base and 117 verbs derived
from a non-prefixed base. If the distribution was not affected by the parameter
prefixed/non-prefixed base, the thirteen verbs derived from a prefixed base
would be distributed as shown in the fourth column of Table 34: there would be
more verbs where the prefix pri- has the submeaning ARRIVE or ATTACH, and
many fewer verbs where the prefix pri- has the submeaning ADD or ATTENUATE.
Table 34: Distribution of the verbs with different bases among the sub-meanings of the prefix
pri- (actual and expected number of verbs)
However, in the RNC the distribution is different. The majority of the verbs with
a prefixed base choose the submeaning ATTENUATE. The distribution of verbs
with prefixed and non-prefixed bases among the submeanings of the prefix pri-
is significantly different from the distribution by chance (P = 4.623E-06, Fisher
test). Therefore the morphological structure of the base significantly affects the
choice of the prefix. 69% of prefixed bases choose the submeaning ATTENUATE.
We can also test the impact of each of the factors comparing the actual distribu-
tion in the corpus with the expected distribution of the same parameter using
the chi-square test. Comparing prefixed base verbs with their expected distribu-
tion shows us significant impact of the factor (χ2 = 30, df = 3, P = 1.3E-06), while
the distribution of the non-prefixed bases does not differ significantly from the
expectation (χ2 = 3.54, df = 3, P = 0.31).
Thus, the submeaning ATTENUATE is frequently used when the base verb is a
prefixed verb. The fact that a Russian verb may have two prefixes simultaneous-
82 | Semantic profiling, predictability and prototypicality
ly has long been noticed in the literature concerned with Russian prefixation
(see Ramchand 2004, Romanova 2004, Svenonius 2004a and references there).
Ramchand, Romanova and Svenonius propose that that the prefix that is closer
to the base functions as a “lexical” prefix, while the second prefix added to the
prefixed stem functions as a “superlexical” prefix. The lexical prefix has a spa-
tial meaning, while the superlexical prefix has an actional meaning. Actional or
Aktionsart meaning is a property of a predicate that refers to the internal tem-
poral structure of the situation (cf. Bache 1985: 10). ATTENUATE is the only ac-
tional meaning available for the prefix pri-, since it describes temporal structure
of the situation, while other submeanings, ARRIVE, ATTACH and ADD describe
spatial configuration of the situation. The distribution of the ATTENUATE sub-
meaning supports the hypothesis of “lexical” vs. “superlexical” distribution of
prefixes: the verbs that are formed from a prefixed base most frequently have
the submeaning ATTENUATE.
However material in this chapter also shows that the actional meaning of
the prefix pri- added to a prefixed base is a tendency rather than a rule. Verbs
with a frequency of more than 100 include thirteen verbs derived from a pre-
fixed base. Nine of them have the submeaning ATTENUATE, while the remaining
four are distributed among other submeanings. See more data on the associa-
tion between the submeaning ATTENUATE and prefixed base among infrequent
pri-verbs in section 4.2.3.
Let us turn to the semantic tags of the verbs. The distribution of the verbs of
the class “move” is significantly different from the chance distribution; the
probability that this distribution occurred due to chance is P = 2.489E-05 (Fisher
test). The semantic tag “move” has strong preference for the submeaning AR-
RIVE. This preference is natural since ARRIVE is the submeaning mostly realized
with motion verbs. Among all imperfective verbs that produce perfectives with
prefix pri- there are twenty-six verbs that belong to the semantic tag “move”. In
nineteen of them the prefix pri- has the submeaning ARRIVE, therefore the tag
“move” predicts the submeaning ARRIVE in 73% of all verbs.
Table 35: Distribution of verbs of movement among the submeanings of the prefix pri- (num-
ber of verbs)
The semantic tag “impact” also affects the distribution of the verbs among sub-
classes. The probability of the outcome of the observed data in Table 36 due to
chance is P=3.332E-06 (Fisher test). While the “move” semantic tag has a pref-
erence for the submeaning ARRIVE, the “impact” semantic tag has an equally
strong preference for the submeaning ATTACH. This attraction is also under-
standable since impact usually involves direct contact. Out of twenty-four verbs
attested with the “impact” semantic tag, eighteen choose the submeaning AT-
TACH, so the “impact” tag predicts the submeaning ATTACH for 75% of the verbs.
Table 36: Distribution of verbs of impact among the submeanings of the prefix pri- (number of
verbs)
The “speech” semantic tag is one of the smallest tags among the imperfective
verbs that produce perfectives with the prefix pri-: there are only six verbs with
this tag. Yet, the distribution of the submeanings among these verbs is affected
significantly (P = 0.0276, Fisher test). The most popular submeaning here is
ADD. It is the submeaning attested for three out of six verbs, in addition to two
verbs of the subclass ARRIVE, and one verb of the subclass ATTENUATE. Thus the
“speech” tag predicts the submeaning ADD and this prediction is successful for
50% of the verbs. However since the tag is associated with so few verbs in this
study, there is not enough data give a meaningful explanation for this prefer-
ence.
Table 37: Distribution of speech verbs among the submeanings of the prefix pri- (number of
verbs)
The “creation” semantic tag is also infrequent in this data; only nine verbs with
this tag are present among the imperfective verbs that produce perfectives using
the prefix pri-. Nevertheless, this tag significantly affects the distribution among
the submeanings (P = 0.0122, Fisher test). The “creation” semantic tag predicts
84 | Semantic profiling, predictability and prototypicality
the submeaning ATTACH, and this prediction is successful for seven verbs out of
nine, i.e. for 78% of the verbs. This attraction is also natural since creation gen-
erally involves contact.
Table 38: Distribution of verbs of creation among the submeanings of the prefix pri- (number
of verbs)
Table 39: Distribution of noncausative verbs among the submeanings of the prefix pri- (num-
ber of verbs)
The distribution of the verbs with the semantic tag “сausative” among the sub-
meanings of the prefix pri- is not significantly different from the distribution by
chance (P = 0.0749, Fisher test). As a result this semantic tag is excluded from
the model.
Table 40: Distribution of verbs of causation among the submeanings of the prefix pri- (number
of verbs)
Table 41 shows what semantic tag predicts what submeaning (first and second
column of Table 41), how many verbs are correctly predicted (third column),
how many relevant verbs are attested overall for each semantic tag and what
percent of the all attested verbs receive a correctly assigned prefix submeaning
(last column).
If our pri-robot simply applies the rules listed in (38), it receives the results de-
scribed in Table 42. For fifty-two verbs (40%), the rules predict one submeaning
of the prefix and this is a correct prediction. Twenty-five verbs (19%) receive one
or several submeaning predictions, however for these verbs all predictions are
incorrect. Thirteen verbs receive two or more conflicting predictions, however
one of them is a correct prediction, so such cases can be resolved by ranking the
rules. Finally forty verbs (31%) do not have a semantic tag in the RNC, and
therefore for submeaning of the prefix for these verbs cannot be predicted using
these rules.
pref ⇒ ATTENUATE
move ⇒ ARRIVE
impact ⇒ATTACH
create ⇒ ATTACH
noncaus ⇒ ARRIVE
speech ⇒ ADD
Table 42: Distribution of the results based on predictions given by semantic and derivational
tags
Let us consider the forty verbs in the dataset that do not have any of the seman-
tic tags investigated in this study and therefore do not receive any prediction.
The distribution of these verbs among the subclasses of the prefix pri- is given
in Table 43. As can be seen most frequently, in nineteen cases out of forty, un-
annotated verbs belong to the subclass ATTACH. In terms of our pri-robot, the
results can be significantly improved if we add a default rule: If the verb does
not have any marking in terms of the semantic tag, assume it has the submean-
ing ATTACH. Using the default rule, the robot correctly predicts the submeaning
for nineteen out of forty unmarked verbs.
subclass verbs
ARRIVE 10
ATTACH 19
ADD 8
ATTENUATE 3
Total 40
For seven verbs derivational and semantic tags of the base verb give different
predictions. In these cases the prediction of the derivational tag “pref” gives
more correct predictions, therefore robot’s results are improved if the robot
always ranks the derivational tag “pref” as more important than any semantic
tag. This tells us that in assigning the submeaning of the prefix morphological
structure is more important than the semantic tag of a verb.
Thirteen out of 117 verbs with a non-prefixed base have more than one se-
mantic tag. In some cases these semantic tags give the same prediction. For
example, vjazat’ ‘knit’ has both “impact” and “create” tags. Both semantic tags
predict that the verb choses the submeaning ATTACH, and this is the correct pre-
diction. However for six verbs the predictions given by the semantic tags that
they belong to are different. For example, the verb drat’sja ‘fight’ has the “im-
pact” and “noncausative” semantic tags. The “impact” tag predicts that the
verb chooses the submeaning ATTACH, while the “noncausative” tag predicts the
submeaning ARRIVE. In this case one of the submeanings, namely ATTACH, is the
correct prediction. In such cases the performance of the pri-robot may be im-
proved if we order the applied rules in the robot’s system. The rule ranking
Predictability of the submeaning | 87
impact>noncaus states that in the case of the presence of both semantic mark-
ings the semantic tag “impact” is more important for the prediction of the sub-
meaning, so its prediction overrules the prediction of the semantic tag “non-
causative”. The rule rankings move>speech, impact>noncaus and
move>noncaus improve the robot’s predictions by one verb each. The rule
speech>noncaus improves the predictions for three verbs, but it makes one
incorrect prediction. Thus the ordering speech>noncaus improves three predic-
tions out of four, while the opposite ordering would improve only one predic-
tion, so the robot’s predictions are still improved if we include this rule in the
robot’s system. The results of the rules for conflict resolution are summarized in
the Table 44.
Even though rules of conflict resolution are based on extremely limited materi-
al, they offer interesting insights into the relative importance of the semantic
tags. The order of the semantic tags presented in (39) shows that the most ab-
stract semantic tag “noncausative” has the least effect on the submeaning of
the prefix, while the most concrete semantic tag “move” has the most effect,
leaving tags “speech” and “impact” in between. Thus, we can see that the rules
in (39) are in agreement with the usual assumptions about the core and the
periphery of semantic classification.
Now the robot has simple rules like “Semantic tag “move” predicts the sub-
meaning ARRIVE,” plus a default rule predicting what submeaning to choose
when no semantic tag is assigned. For the base verbs that are marked with sev-
eral tags, which predict different results, the robot uses a ranked hierarchy of
rules.
Application of all the rules allows the robot to predict the submeaning cor-
rectly for eighty-one verbs out of a total 130, which is 62% of the verbs; see Ta-
ble 45.
88 | Semantic profiling, predictability and prototypicality
Even though the robot is not able to predict the submeanings for all verbs, its
results give us linguistically relevant information. As mentioned before, within
cognitive linguistics compositionality is viewed as a scale, where every item is
placed either closer to full compositionality or closer to total indivisibility. The
robot’s result places the prefix pri- on the scale of compositionality and even
gives precise measurement as to where on the scale of compositionality the
prefix pri- is positioned. Now we can answer that according to the result of this
study the Russian prefix pri- is 62% compositional.
The pri-robot has analyzed the frequent verbs and has constructed rules based
on them. This section describes how well the robot’s rules apply to verbs unfa-
miliar to the robot. For this section I use the verbs that have a frequency less
than 100 in the modern subcorpus of the RNC. Predicting the meaning of the
infrequent items of the class using the meanings of frequent representatives of
the class (in our case the submeanings of the prefix pri-) models the process of
language acquisition. The child acquiring a language is more exposed to the
frequent representatives of the class, and based on them he or she acquires the
schema for the class itself. The acquired schema is then applied to the infre-
quent representatives of the class, see for example Tomasello 2003.
There are altogether 526 infrequent verbs with the prefix pri-. 228 of them
are marked for semantic tags in the RNC. Only 202 of the verbs belong to one of
the semantic tags that are in the list that the robot operates on: “move”, “im-
pact”, “create”, “speech”, “noncaus”. It would be unfair to the robot to ask it to
analyze verbs that are not marked for the semantic tags it knows, so the robot is
tested only on the verbs that are marked for one of the semantic tags that it
recognizes. The robot, using rules listed in (38) and rules of conflict resolution
listed in (39), correctly predicts the submeaning of pri- for exactly one half of
the infrequent pri-verbs: 101 verbs out of 202; see Table 46.
Predictability of the submeaning | 89
Table 46: The number of correct results based on predictions for infrequent verbs
We can see that for the infrequent verbs the robot is making correct predictions
only for 50% of the verbs, while for the frequent verbs the percent of the robot’s
correct predictions is 62%. The difference between the correct result for the
frequent and infrequent verbs in Table 47 is statistically significant (χ2= 10.43, df
= 1, P = 0.0012).
Why is the robot making worse predictions for the infrequent verbs? This differ-
ence can be explained by the different distribution among submeanings for
frequent and infrequent verbs. Table 48 and Figure 10 show the distribution of
the submeanings among the infrequent verbs that are marked for the semantic
tags available to the robot.
Table 48: The distribution of the submeanings of the prefix among the infrequent pri-verbs
marked for one of the semantic tags available to the robot
Fig. 10: The distribution of the submeanings of the prefix among the infrequent pri-verbs
marked for one of the semantic tags available to the robot
It is important to note that 202 infrequent verbs with semantic tags have the
same distribution of submeanings as all 526 infrequent verbs with the prefix
pri-; see Table 49 and Figure 11.
Table 49: The distribution of the submeanings of the prefix among the all infrequent pri-verbs
Fig. 11: The distribution of the submeanings of the prefix among the all infrequent pri-verbs
Predictability of the submeaning | 91
We can see that for the infrequent verbs the dominant submeaning is ATTENUATE.
However for the frequent verbs the submeanings ATTACH and ARRIVE are domi-
nant; see Table 50 and Figure 12.
Table 50: The distribution of the submeanings of the prefix among the frequent pri-verbs
Fig. 12: : The distribution of the submeanings of the prefix among the frequent pri-verbs
The distribution of the submeanings of the prefix pri- among frequent and in-
frequent verbs show several similarities, for example, the submeanings ARRIVE
and ATTACH are used for almost the same number of verbs (46 and 47 for the
frequent verbs, 127 and 128 for the infrequent verbs), so we can conclude that,
first, these two submeanings are equally distributed among frequent and infre-
quent verbs and, second, these two meanings have equal importance in the
radial category of the prefix pri-. However, the different impact of the submean-
ing ATTENUATE makes the distribution of the submeaning among frequent verbs
and infrequent verbs significantly different (χ2 = 24.69, df = 3, P < 0.0001). So we
have to conclude that native speakers of Russian, when deriving frequent and
infrequent verbs with the prefix pri-, use two different strategies. Low frequency
of the derived verb with the prefix pri- triggers the submeaning ATTENUATE to be
more salient, while the submeanings ARRIVE and ATTACH are more salient for the
frequent verbs.
92 | Semantic profiling, predictability and prototypicality
Table 51: The distribution of the submeanings of the prefix among the infrequent pri-verbs
produced from a prefixed base
subclass verbs
ARRIVE 6
ATTACH 2
ADD 2
ATTENUATE 73
The difference in the distribution of the submeanings for the frequent and in-
frequent verbs of Russian explains why the robot’s predictions worsen when the
robot analyzes the infrequent verbs. However further studies are needed to
explore the question from the area of language acquisition: how do children
who are mostly exposed to the frequent verbs, acquire the salience for the AT-
TENUATE submeaning, which is so pronounced for the infrequent verbs.
Summing up the pri-robot experiment, we can say that the pri-robot provid-
ed with derivational and semantic tags of verbs is able to correctly predict the
submeanings of the prefix pri- in 62% of the frequent verbs and 50% of the in-
frequent verbs. If the robot chose among the four submeanings by chance, it
would have predicted the submeaning of the prefix correctly only in 25% of the
verbs. This shows that the semantics of a base verb is strongly correlated with
the submeaning of the prefix. However based on these results we cannot claim
that the prefix pri- in Russian is not polysemous since the submeaning of the
prefix cannot be fully predicted based on the meaning of a base verb. This dis-
tinction is caused by a different distribution of the submeanings for frequent
and infrequent verbs. While frequent verbs contain more verbs with the sub-
meanings ARRIVE and ATTACH, the infrequent verbs have more verbs with the
submeaning ATTENUATE. Moreover the infrequent verbs contain significantly
The quest for the prototype | 93
This section departs from the computational model and explores how the radial
category of the prefix pri- may be categorized in the minds of native speakers of
Russian. In order to uncover the internal structure in the radial category, i.e.
determine which are central and non-central elements of the category, we need
to discover what the prototype of the prefix pri- is.
This list contains more features than are traditionally listed. For example
Gilquin (2006: 180) who also ends her discussion of prototypicality with a set of
features only lists four, which are summarized in features 2, 4, 5, and 6 of the
list above. Even though the list compiled above presents more features con-
structing what a prototype is it is hard not to agree with the Gilquin’s conclu-
sion: “The various facets of prototypically can converge, when all point in the
same direction, but they can also be (wholly or partly) divergent and reflect
different realities. In the former case, the prototype may be said to be more
prototypical than in the latter case.” Each of the characteristics of the prototype
is discussed in more detail below.
Let us follow the list of the features of the prototype and look at which sub-
meaning has more prototypical features. First on our list is semantic of center of
gravity of the radial category.
An item is identified as the center of gravity of a radial category if more
nodes in the network are connected to it than to any other node. This means
that this item is connected to the most nodes in the category via metaphor or
metonymy. Such an item serves as a prototype in the sense that we produce
other items from this most basic meaning. Figure 13 here repeats Figure 6 from
section 4.1 presenting the radial category for the prefix pri-. We see that this
radial category is balanced in that every node is connected with two other
nodes: ARRIVE is related to ATTACH because attaching means the endpoint of
arrival; ATTACH is related to ADD because adding can be seen as attaching small
amount of substance to larger amount of substance; ADD is related to ATTENUATE,
because ATTENUATE can be viewed as adding a smaller amount of event to larger
amount of an event; and finally, ATTENUATE is related to ARRIVE because both
ATTENUATE and ARRIVE are applicable to events, where ATTENUATE describes mov-
ing slightly towards the endpoint of the event, while ARRIVE describes reaching
the endpoint. Thus, all elements of the radial category of the prefix pri- are
equal in terms of being the center of gravity of the network. Thus, this criterion
does not help us to decide which meaning is the prototype.
The quest for the prototype | 95
Let us now turn to the number of characteristic features of the category. The
item with the most relevant characteristics can be thought of as being closest to
the overall schema. If we are thinking about the concept of mother in terms of
biological, social and psychological models, the most prototypical mother
would be a mother who biologically, socially and psychologically functions as a
mother. Non-prototypical examples such as surrogate mother, or adoptive
mother would be considered mothers according to only one or two characteris-
tics. A surrogate mother is a mother only according to a biological model, but
not according to a social or a psychological model, while an adoptive mother is
a mother according to social and psychological models, but not according to a
biological model. Thus, the item that has the most category features serves as
the prototype.
The semantic schema for the prefix pri- is ‘approaching’ (see Endresen et al.
2012), which is most closely realized in two submeanings: ARRIVE and ATTACH.
These two submeanings show direct realization of the schema. Both arrival and
attachment presuppose direct approach. The two other meanings are less di-
rectly connected to the schema. However it can be argued that for the ADD
meaning, the smaller amount of substance is approaching the larger amount of
substance, as well as for the ATTENUATE meaning, the smaller amount of an
event is approaching the resulting event. Still both these meanings are further
from the initial ‘approaching’ schema than ARRIVE and ATTACH. Thus, character-
istic features of the category point to ARRIVE and ATTACH as prototypical sub-
meanings.
It is agreed that the prototype is usually concrete rather than abstract, since
it is more natural for people to analyze concrete items as simple, and describe
more abstract items in terms of concrete items. This mechanism underlies the
idea of metaphor, which provides a cross-domain mapping between two con-
96 | Semantic profiling, predictability and prototypicality
Let us now turn to entrenchment. The term “entrenchment” refers to how avail-
able in memory linguistic items are. While some items are used rarely and re-
quire derivation each time they are used, other items are needed frequently and
for easier access such items are stored in memory in prepackaged format.
“[F]amiliar concepts like ‘dog’ or ‘poodle’ are deeply entrenched (italized by
Schmid) in our memory so that their activation has become a highly automated
routine” (Schmid 2007: 118). According to Langacker (1987: 59), there is a
tem.” Gilquin (2006: 168) notes that “[g]iven the vagueness surrounding the
term “prototype”, as well as the complexity involved in testing linguistic proto-
typically experimentally (how does one get people to judge the “goodness-of-
example” of, say, a particular transitive clause?), it comes as no surprise that
frequency in linguistic usage has regularly been used as a methodological
short-cut to establish the prototype.”
However in more recent studies it has been observed that there are some in-
consistencies between frequency, salience and entrenchment. Tsohatzidis
(1990: 8) points out that the “undeniable heuristic value of the notion of proto-
typicality should not obscure the fact that its exact theoretical shape is less
clear than one might have wished, especially when it is transferred from purely
psychological to specifically linguistic domains of investigation.” This insight is
supported by several studies comparing results of elicitation and corpus analy-
sis, which reveal that the items getting highest scores in the elicitation tests do
not always coincide with the most frequent items (Sinclair 1991, Geerarts, Grin-
delaers and Bokema 1994, Aitchison 1998, Roland and Jurafsky 2002, Gilquin
2006, Schmid 2007).
Since we do not have access to experimental data in this study, entrench-
ment will be studied through frequency, as its available correlate. However we
need to consider the question of whether we need to measure token or type
frequency. It is well known that some linguistic phenomena are sensitive to
token and some to type frequency. Token frequency refers to how often a lin-
guistic unit is found overall and type frequency refers to how many different
types of words or constructions the unit is found in. For example, for the sub-
meanings of the prefix pri-, token frequency determines how often each sub-
meaning is found in the corpus – how many occurrences of each submeaning
are found in the corpus. By contrast, type frequency refers to how many differ-
ent verbs are associated with each submeaning. Let us consider both type and
token frequency of submeanings of the prefix pri- in the following discussion.
In terms of token frequency the most frequent submeaning is ARRIVE; see
Table 52 and Figure 14. It used in 71% of all verb attestations. This means that
out of any ten examples of uses of the prefix pri- that a speaker of Russian en-
counters, in seven occurrences the prefix pri- means ARRIVE. This is also im-
portant information if we consider how Russian children master the meaning of
the prefix pri-. Since most often they would encounter the submeaning ARRIVE,
this should be the meaning that children acquire first for this prefix.
98 | Semantic profiling, predictability and prototypicality
subclass verbs
ARRIVE 231,251
ATTACH 55,014
ADD 30,729
ATTENUATE 17,027
If we explore the type frequency of the submeanings, we see that the most fre-
quent submeaning here is ATTENUATE; see Table 53 and Figure 15. There are
more verbs that have the submeaning ATTENUATE than verbs with any other
submeaning.
subclass verbs
ARRIVE 173
ATTACH 175
ADD 76
ATTENUATE 206
The quest for the prototype | 99
This difference is due to the fact mentioned in the previous section that frequent
and infrequent verbs with the prefix pri- have different distributions in terms of
the submeanings of the prefix pri-. This difference is notable for both token
frequency and type frequency.
Table 54: Distribution of submeanings for frequent and infrequent verbs: token frequency
Fig. 16: Distribution of submeanings for frequent (left) and infrequent (right) verbs: token
frequency
100 | Semantic profiling, predictability and prototypicality
If we look at token frequency, we see that the distribution for frequent verbs
repeats the overall distribution, cf. Table 54 and left chart of Figure 16. The
submeaning ARRIVE is used for 75% of the frequent verbs (it is used for 71% of all
verbs). However the distribution for infrequent verbs is different. The submean-
ing ARRIVE is used for 21% of the infrequent verbs, while the submeaning ATTEN-
UATE is used for 30% of the verbs, and the submeaning ATTACH is used for 37% of
verbs.
If we compare the type frequency of the verbs that have more than 100 oc-
currences and less than 100 occurrences in the RNC10, we note that the sub-
meaning ATTENUATE is poorly represented in the group of frequent verbs (only 19
verbs as opposed to 46 of ARRIVE), but predominates among the infrequent verbs
(187 verbs as opposed to 127 of ARRIVE).
Table 55: Distribution of submeanings for frequent and infrequent verbs: type frequency
Fig. 17: Distribution of submeanings for frequent (left) and infrequent (right) verbs: type fre-
quency
Recent works have shown that “forms of high token frequency will be more
autonomous and more likely to be unanalyzed and less likely to participate in
schemas; high token frequency forms will thus not contribute to the productivi-
ty of the pattern” (Bybee 2007: 174). Verbs with higher token frequency most
often occur with the submeaning ARRIVE. For example, let us consider verbs that
||
10 Table 55 and Figure 17 repeat Tables 49-50 and Figures 11-12 from section 4.2.3.
The quest for the prototype | 101
have token frequency more than 1000 in the RNC. Out of thirty such verbs,
twenty-one have the submeaning ARRIVE (the remaining verbs are distributed
evenly: 3 verbs with the submeaning ATTACH, 3 verbs with the submeaning ADD,
3 verbs with the submeaning ATTENUATE). Therefore, the verbs with high token
frequency that use the submeaning ARRIVE are likely to be unanalyzed and be
used as “prefabricated” units. At the same time “[f]orms with lower token fre-
quency will be learned more easily if they can be related to other stored forms”
(ibid.). In our example the submeaning ATTENUATE is mostly used with verbs of
low token frequency, so this submeaning is more likely to be seen as produc-
tive. Thus, the submeaning ATTENUATE shows two features correlating with
productivity. It has the highest type frequency and it is mostly used with verbs
of low token frequency. Hence, we can conclude that token frequency and type
frequency suggest that different members of the radial category of the prefix pri-
are prototypical. Token frequency points to the submeaning ARRIVE – it occurs in
seven out of ten uses of the prefix pri-. However type frequency and productivity
point to the submeaning ATTENUATE, because it has the highest type frequency
and is the productive meaning of the prefix pri-.
A distribution pattern in which the item that has the highest token frequen-
cy is different from the item that is characterized by the highest type frequency
is not unusual for a linguistic phenomenon. The most famous and mostly thor-
oughly investigated example of such a distribution is the use of past tense end-
ings in English. Here words with the highest token frequency have irregular
past tense forms such as went, came, gave, took and so on, while words of lower
frequency use the past tense ending -ed. While the irregular pattern has a high-
er token frequency, -ed has the highest type frequency and is the productive
pattern for the past tense in English (for more details see Bybee and Slobin 2007
and references therein). Experimental evidence suggests that English irregular
verbs are stored as prefabricated units in the lexicon, while -ed forms of verbs of
lower frequency are derived11. I speculate that the distribution of the submean-
ings of the prefix pri- is similar. Most verbs with the submeaning ARRIVE have
||
11 It is important to note that English is a language with poor inflectional morphology and the
English system contrasts one regular past tense ending with all irregular patterns. Languages
with rich inflectional morphology, where several regular inflectional patterns are available,
show a different picture: regular and irregular conjugation patterns are not opposed, but rather
form a gradient scale as shown in experiments involving acquisition of conjugational inflec-
tions in Italian (Orsolini & Marslen-Wilson1997), German (Clahsen 1999), Icelandic (Ragnasdót-
tir et al. 1996), Norwegian (Simonsen 2000), Finnish (Niemi 2006), and Russian (Gor and Cook
2010). These findings contribute to one of the important topics in this book: gradience of most
linguistic phenomena.
102 | Semantic profiling, predictability and prototypicality
very high token frequency and are stored in the mind of speakers as prefabri-
cated units. The verbs of lower frequency use a regular pattern, i.e. the sub-
meaning ATTENUATE, and are derived using the prefix pri- and the base verb.
Since the distribution of the prefixes is so different for frequent and infrequent
verbs, this suggests that children acquiring the prefix pri- probably go through
several stages. First, they encounter more frequent verbs, which suggest that
pri- is associated with the submeaning ARRIVE. Second, once they acquire
enough less frequent verbs, they reanalyze the structure of the submeanings,
because with more data, they are able to deduce that the submeaning ATTENUATE
is the productive submeaning of the prefix pri-. Acquisition of the prefix pri-
needs further investigation.
Let us now consider salience as a feature of the most prototypical element in the
network. To put it bluntly, an item is the most “salient” representative of a cate-
gory if it is the first that comes to mind. Usually linguists distinguish between
cognitive salience and ontological salience. Cognitive salience refers to the
activation of an item in the memory in a concrete situation. For example, activa-
tion of the word railroad can occur in two different situations: it can be either
because the concept itself is activated, or activation of one concept can trigger
the activation of the other, like in the citation from Lewis Carroll below, when
Alice concludes that she has fallen into the sea, and the concept of the sea trig-
gers for her the concept of railroad connected with it:
As she said these words her foot slipped, and in another moment, splash! she was up to
her chin in salt water. Her first idea was that she had somehow fallen into the sea, “and in
that case I can go back by railway,” she said to herself. (Alice had been to the seaside once
in her life, and had come to the general conclusion, that wherever you go to on the Eng-
lish coast you find a number of bathing machines in the sea, some children digging in the
sand with wooden spades, then a row of lodging houses, and behind them a railway sta-
tion.) (Carroll, Lewis. Alice's Adventures in Wonderland. CHAPTER II. The Pool of Tears)
portraying a person on a beach, the human eye is more likely to distinguish the
person than the beach. It is true that ontologically salient items tend to be cog-
nitively salient, but not vice versa.
The most salient element is the element that comes first to mind in the pro-
cess of elicitation. The most natural method of studying salience is therefore an
elicitation experiment. Since this work is corpus-based, experimental data is
not available. However it is possible to study phenomena that are available in
the corpus and correlate with salience. In this chapter I explore two such corre-
lates for salience for the prefix pri-: 1) the submeaning that is default for the pri-
robot 2) the submeaning appearing with so-called “pro-verbs” (defined below).
4.3.4.2. Pro-verbs
In Russian there are verbal roots such as kljačit’, figačit’, figarit’, sandalit’, so-
bačit’, šibarit’, špandolit’, xrenačit’ that can be characterized with two features:
1) etymologically these roots are usually derived from names of animals and
taboo words, 2) the verbal roots cannot be characterized with separate mean-
ings and their meaning changes depending on the construction used in the
sentence. These verbs stand in place of a verb, similar to how a pronoun usually
stands for a noun (see Raskin 1978, Krongauz 1998). The wide scope of mean-
104 | Semantic profiling, predictability and prototypicality
ings available to such verbs can be illustrated by the range of examples with the
verb figačit’. Since such verbs are highly colloquial and infrequent in standard
speech, corpus data cannot be used to describe their semantics. In this study I
used the first 100 examples given by the Yandex search engine12 for the past
tense form of figačit’: figačil. This distribution gives us some idea of how the
meanings of this verb are distributed overall.
Table 56 shows us that a wide range of meanings is available for the verb
figačit’. This range suggests that this verb does not have its own specific mean-
ing, but functions as a verbal variable, acquiring a meaning from the construc-
tion used in a sentence. Thus we are dealing with a set of pro-verbs, which oc-
cur in a range of contexts that indicates that these verbs do not have specific
meaning and do not even belong to a specific semantic class.
These verbs interest us in the context of the prefix pri- because they are able
to produce prefixed perfectives with the prefix pri-. Since the base verbs in such
perfectives are pro-verbs, they do not imply any choice of the submeaning of
the prefix. Therefore the submeaning used in such contexts should be the de-
fault submeaning of the prefix pri-. The search for the uses of the prefixed verb
form prifigačil in the Yandex search engine shows that 99 out of the first 100
examples belong to the semantic class ‘impact’ and use the submeaning ATTACH
of the prefix pri-. The one remaining example denotes transportation and uses
the submeaning ARRIVE of the prefix pri-. The verb prifigačit’ has a more narrow
scope of meanings than its base, the pro-verb figačit’.
||
12 www.yandex.ru. The search was done in October 2011.
13 Source: http://www.ehome.by/?TopicID=17
14 Source: http://edgun.narod.ru/
15 Source: http://awd.ru/student.htm
The quest for the prototype | 105
From seven different classes present for the verb figačit’, there is left only one
dominant class for the verb prifigačit’. Note that the ‘impact’ class is not present
in Table 56, so it is not one of the frequent meanings for the pro-verb figačit’.
The meaning of prifigačit’ is more affected by the semantics of the prefix pri-
than by meanings available to the pro-verb figačit’. The prefix pri- combined
with the pro-verb figačit’ shows strong preference for the submeaning ATTACH.
Other pro-verbs prefixed with pri- show preference for the same submeaning;
see examples (40)–(43).
(42) I stal uglubljat' jamki dlja stojanov, potom pomog postavit' pleten', pri-
kljačil ego k stojanam i ušel. [M. A. Sholoxov. Tixij Don. Kniga četverta-
ja (1928-1940)]
‘He began to deepen the holes for the poles, then he helped to put up
the fence, attached the poles to it and left.’
��
16 Source: http://www.renault-club.ru/showthread.php?t=16785
17 Source: http://teenslang.su/id/7745
18 Source: http://statusje.com/user/1338/
19 Source: http://jukov.tanais.net/songs-text.htm
106 | Semantic profiling, predictability and prototypicality
We see that the semantic profiles of the prefixed and non-prefixed pro-verbs are
different. Non-prefixed pro-verbs appear in highly frequent constructions, func-
tion as verbal variables and obtain their meaning using the meaning of the
construction. Pro-verbs with the prefix pri- are used to signal attachment. Since
pro-verbs themselves do not specify the meaning of these prefixed forms, the
meaning of the forms reveals the default meaning of the prefix. Hence the pro-
verbs point to the submeaning ATTACH as the most salient submeaning of the
prefix pri-. Therefore two phenomena that correlate with salience, the default
meaning for the pri-robot and the submeaning used with the pro-verbs, indicate
that the submeaning ATTACH shows more prototypical features. .
Summing up the section on the prototype of the prefix pri-, we can conclude
that different prototype features point to different elements of the radial catego-
ry of the prefix pri-. The submeanings ARRIVE and ATTACH show the most charac-
teristic features of the category. The submeaning ARRIVE is used with the motion
verbs, which are claimed to serve as prototypical verbs in the Russian aspectual
system. It may be concluded that native speakers of Russian employ different
strategies for most frequent verbs, least frequent verbs and in a default setting:
the most frequent verbs are associated with the meaning ARRIVE, the least fre-
quent verbs are associated with the meaning ATTENUATE, and when the meaning
of the verb is not specified, the submeaning ATTACH is used. Our understanding
of the term prototype does not include a possibility for all these features to point
to different elements, which means that we need to adjust our understanding of
the prototype to account for these findings.
4.4. Summary
The submeanings of the prefix pri- are not fully predictable. The pri-robot – a
model based on semantic profiling – uses rules and the derivational and seman-
tic structure of the base verbs and is able to predict submeanings for 62% of
frequent verbs and 50% of infrequent verbs. This shows that the distribution of
submeanings is not due to chance and that semantic profiling gives reasonable
results in predicting the submeaning. However the distribution of the submean-
ings of the prefix pri- does not make full predictability possible.
||
20 Source: http://www.virtualireland.ru/archive/index.php/t-37603.html
Summary | 107
General solutions are always preferred to more specific ones, and it is assumed that any
regularities that can be captured by means of general rules and principles should be so
captured. It is also held as a matter of course that anything that is computable need not
be, and hence is not, stored. However, it is doubtful that the principle of economy is an
appropriate standard by which to evaluate linguistic theories, if these are intended to be
psychologically realistic. Human brains are relatively slow processors, but have enormous
storage capacity. From a psychological point of view, therefore, retrieval from memory is
the preferred strategy. People store vast numbers of prefabricated units, and rely on them
whenever possible – even when they are fully compositional and could also be assembled
using rules.
The aspectual system of Russian contains two aspects: perfective and imperfec-
tive. Roughly speaking, their meanings can be described as ‘a completed activi-
ty’ for perfective and ‘incomplete, ongoing, habitual, reversed or repeated activ-
ity’ for imperfective. For most lexical verbs the aspect is assigned to a verb, so
every verb is either perfective or imperfective. There are several hundred so-
called biaspectual verbs (velet’ ‘order’, ranit’ ‘injure’, naputstvovat’ ‘express
one’s best wishes’, sodejstvovat’ ‘assist’, etc.). However for every verb used in a
sentence a native speaker of Russian is able to determine unequivocally which
of the two aspects is used in the sentence. For the biaspectual verbs, aspect can
be viewed as syncretic, similar to number in English words like fish, sheep, deer.
Criteria for establishing an aspectual pair | 109
��
21 Элементами одной парадигмы (= формами одной лексемы) могут более или менее
ненасильственно считаться лишь пары типа упаковать-упаковывать” – translated from
Russian by J.K.
22 Cited using the second edition. First published in 1948.
110 | Constructional profiling and aspectual pairs
verbs are used in imperfective forms. Maslov’s criterion uses a link between
sentences like those in (44) and (45). Both sentences describe the same event in
the past, however while the former sentence is in past tense, the latter uses the
praesens historicum. The parallelism between these two sentences allows us to
establish that the verbs otkryt’ ‘openPFV’ and otkryvat’ ‘openIPFV’ form an aspectual
pair.
Thus comparing a depiction of the same event described in the past tense and in
praesens historicum we receive perfective-imperfective analogues that differ
only in their aspect. After Maslov, several additional criteria were introduced.
Zalizniak and Shmelev (1997: 42) use a habitual context as an additional criteri-
on, which can be applied for some verbs for which praesens historicum uses are
not possible. Hence, according to Zalizniak and Shmelev, parallelism between
examples (44) and (46) also give us ground for establishing an aspectual pair
otkryt’ ‘openPFV’ and otkryvat’ ‘openIPFV’.
Zalizniak and Shmelev (ibid.) explain why they introduce the habitual criterion
in addition to Maslov’s criterion:
Some events due to their nature cannot be viewed as repetitive (for example, rodit’sja [‘to
be born’] or umeret’ [‘die’] in a literal sense and with a single subject), while others cannot
be presented as occurring “before someone’s very eyes”: for example, it is hard to trans-
form into praesens historicum the sentence Za gody, prošedšie so dnja ix poslednej vstreči,
ee čerty sterlis’ u nego iz pamjati [‘During the years that have passed since the day they last
met, her features have been effaced from his memory’]. Therefore in certain cases using
only one of the criteria does not give a researcher sufficient results 23.
��
23 “[Н]екоторые события по своей природе не могут быть представлены как
повторяющиеся (например, родиться или умереть в буквальном смысле и с единичным
субъектом), а некоторые другие – наоборот, как происходящие «как бы на глазах»: так,
Criteria for establishing an aspectual pair | 111
Bulygina and Šmelev (1997: 102-104) discuss a similar criterion using an impera-
tive context and contrasting it with imperative under negation. According to
this criterion we establish that otkryt’ ‘openPFV’ and otkryvat’ ‘openIPFV’ form a pair
because the former is used in the positive imperative sentence, while the latter
is used in the negative imperative sentence.
Thus, all criteria for establishing an aspectual pair are based on two parallel
syntactic contexts. One of the contexts usually employs an imperfective verb,
while the other usually engages the perfective verb. These two contexts can be
to some extent claimed to describe the same situation and therefore to express
the same meaning.
However, the criteria for establishing aspectual pairs are inconsistent and
raise many problems. These problems can be organized into three groups: 1)
problems related to the formulation of the criteria, 2) problems concerning the
use of criteria 3) problems with relating the criteria to the cognitive process in
the mind of a native speaker. In the next sections I discuss these problems in
detail.
The first and most important problem concerns the exact formulation of the
criteria, which is not usually questioned. There seem to be two possible versions
of how the criteria can be formulated; these two possible versions can be re-
ferred to as universal and existential versions. These terms refer to the differ-
ence between universal and existential quantifiers which is similar to the differ-
ence between the two versions of the criteria. Let me explain the difference
��
например, трудно перевести в praesens historicum фразу За годы, прошедшие со дня их
последней встречи, ее черты стерлись у него из памяти. Поэтому в некоторых случаях
применение только одного теста не дает достаточно определенных результатов” –
translated from Russian by J.K.
112 | Constructional profiling and aspectual pairs
between the two using the praesens historicum criterion as an example. The
universal version of the criterion can be formulated as in (49):
The existential version of the same criterion can be stated as follows in (50):
(50) If there exists one example where a perfective in past tense is inter-
changeable with an imperfective in the praesens historicum, then the
perfective and the imperfective are aspectual counterparts of each oth-
er.
We can see that the two versions, the universal and the existential, are very
different from each other. While the universal version is a very strong claim,
which makes a statement about all possible contexts, the existential can be seen
as a very weak claim about only one context. Note that (49) makes a statement
about all perfective past tense and all imperfective praesens historicum con-
texts, while some scholars claim that in establishing an aspectual pair one
should only go from a perfective to an imperfective (see Zalizniak and Shmelev
2000: 79, Mikaelian, Shmelev and Zalizniak 2007: 318). In this case the univer-
sal version can be reformulated in a milder variant (51):
(51) If for every example a perfective in past tense can be replaced with an
imperfective in the praesens historicum, then the perfective and the
imperfective are aspectual counterparts of each other.
However both variants of the criteria are not sufficient for establishing an aspec-
tual pair. Both variants of the universal version are too strong. For most pairs of
verbs explored in this study, there exists at least one example of the use of the
perfective verb which cannot be replaced by an imperfective, see examples
below that illustrate this:
On the other hand the existential version is too weak, because a single example
where a perfective and an imperfective are interchangeable might indicate a
pair of verbs which we do not want to call a pair. Percov (2001: 127) points out
that such parallelism may occur for different reasons. For example, sentence
(56) with the prefix pere- meaning ‘seriatim’ can be seen as parallel to both
sentences (57) and (58), because of the quantifier adjective ‘all’ that also induc-
es the implication that all objects are affected. However no one would consider
celovat’ ‘kiss’-perecelovat’ ‘kiss all (seriatim)’ a true aspectual pair due to the
clear difference in meaning. Still we see that this difference in meaning can be
compensated by additional elements in the sentence with the same connota-
tion.
(58) Kazdyj raz, kogda on prixodt, on vsex našix devušek smelo celuet.
‘Every time he comes he kisses all our girls (seriatim)’.
114 | Constructional profiling and aspectual pairs
Percov (ibid.: 128) offers the following solution to this problem: “The possible
relations of co-paired sentences have to be restricted in order to avoid false
correlates like celovat’-perecelovat’... we might require that these sentences
could only contain finite forms of verbs, their arguments and possibly some
types of adverbials (for example for the iterative criterion it will be possible to
use such adverbials as vsegda, často, každyj raz etc. [‘always’, ‘often’, ‘every
time’] )”24.
This formulation, even though it improves the original definition, still has
problems with the relevancy of the examples that are compared (this problem is
discussed in section 5.1.2.1) and with cognitive relations between the members
of an aspectual pair (discussed in section 5.1.3).
Most researchers do not differentiate between the universal and existential
versions. They tacitly assume that the lexical features of a verb make it possible
to claim that if one example shows interchangeability of perfective and imper-
fective, then all examples will be interchangeable. This assumption makes the
presupposition that a verb as a lexeme shows rather unified behavior. However
if we turn to corpus examples we will soon see that each verb participates in
several different constructions, and those constructions also add meaning to the
resulting sentence. For example a construction with the universal quantifier ves’
‘all’ has contributed to sentences (56)-(58) and affected the interchangeability of
perfective and imperfective verbs in those sentences. Thus, the final formula-
tion of the criteria must be construction-based, not verb-based. I offer such a
formulation in section 5.1.4.
��
24 “Cледует как-то ограничить допустимые соотношения фраз-партнеров – во
избежание ложных коррелятов типа перецеловать ~ целовать... можно, например,
потребовать, чтобы эти фразы могли содержать только личные глаголы, их актанты, и,
возможно, некоторые типы адвербиальных компонентов (например, для итеративного
подкритерия можно допустить наличие во фразе адвербиалов всегда, часто, каждый
раз и т. п.)” – translated from Russian by J.K.
Criteria for establishing an aspectual pair | 115
5.1.2.1. Relevancy
All the proposed diagnostic contexts rely on a parallelism between two con-
texts. The same situation is described using different strategies. In one of the
strategies (past tense, positive imperative) the perfective verb is usually used,
while the other strategy (praesens historicum, habitual context, imperative with
negation) only allows imperfective verbs. So if we describe the same event using
the second strategy we will get “the same verb, only in imperfective aspect”25
(Mikaelian, Shmelev and Zalizniak 2010: 6), because the second strategy auto-
matically transforms the verb into an imperfective. Since the two contexts refer
to the same situation, we can claim that the perfective and the imperfective
verbs express the same meaning. Note that for Maslov’s original criterion this
seems to be more or less true – praesens historicum is indeed used to describe
one event in the past and differs from the past tense mostly in expressivity.
However the other criteria do not follow this restriction precisely. The criterion
that uses the habitual mode compares one event in the past with a series of
events, so we cannot exactly claim that these two contexts describe the same
situation. The imperative criterion steps even further away from equivalence of
expression: in this criterion we compare an imperative context with its exact
opposite – negation of imperative, so here the two contexts never can be used to
describe the same intention26 in the same situation. We can only observe that
the two intentions behind the positive and negative imperative are somehow
parallel, given the fact that one of them can be produced from the other more or
less automatically. Thus, the parallelism between the two contexts used in the
criteria is questionable and not always clear.
��
25 “тот же глагол, только в несов. виде” – translated from Russian by J.K.
26 Since we are talking about an imperative, which expresses the intention of the speaker, we
cannot say that this context describes a situation, however the intentions behind the positive
and negative imperatives are clearly different.
116 | Constructional profiling and aspectual pairs
ductive than previously assumed, see Kuznetsova and Sokolova (to appear). For
instance, a simplex imperfective goret’ ‘burnIPFV1’ and a secondary imperfective
sgorat’ ‘burnIPFV2’ produced using the prefix s- and the suffix -a- can both be
considered to be aspectual partners of the perfective sgoret’ ‘burnPFV’. In such
triplets the simplex imperfective as goret’ ‘burnIPFV1’ is more likely to appear in
imperative with negation, as in (59a); compare with ungrammatical sgorat’
‘burnIPFV2’ in the same context (59b). While the secondary imperfective of the
triplet sgorat’ shows preference for praesens historicum and habitual contexts,
see (60a), where the simplex goret’ is ungrammatical in similar contexts, see
(60b). This phenomenon is described in more detail in Janda et al. (2013), Kuz-
netsova and Sokolova (to appear).
It is doubtful that we can use syntactic criteria for establishing aspectual pairs if
the criteria themselves do not agree with each other. It has to be noted that this
is a frequent problem for phenomena based on syntactic criteria, cf. mismatches
in the proposed diagnostics for unaccusativity (see Alexiadou, Anagnostoup-
oulou and Everaert 2004).
Last but not least, a problem with the use of the criteria for establishing an as-
pectual pair is the fact that linguists who are using the criteria get very different
results. Thirty-six linguists participated in a survey conducted at an aspectual
seminar at Moscow State University (see Čertkova et al. 1997). In this survey the
participants were asked to consider several pairs of morphologically related
Criteria for establishing an aspectual pair | 117
verbs and evaluate them as aspectual pairs. Nine of the thirty-six participants
mentioned that they used Maslov’s criterion to distinguish the aspectual pairs.
Table 57 below shows their judgments for seven pairs of morphologically relat-
ed verbs.
Table 57: Aspectual relations between the seven pairs of verbs as judged in a survey (Anketa
1997) using Maslov’s criterion
Table 57 is structured as follows. The first column shows the pair of morpholog-
ically related verbs in question. The second column provides the gloss for the
pair. The third column shows how many linguists using Maslov’s criterion con-
sidered this pair to be an aspectual pair. The fourth column shows for how
many linguists this pair is clearly not an aspectual pair. The last column gives
information on how many linguists agreed that these verbs may be seen as a
pair but only with certain restrictions and reservations. The first pair prygat’-
prygnut’ ‘jump’ was judged by all nine participants, and the other six pairs were
judged only by eight participants each. As we can see, even linguists, who
themselves claimed that they used Maslov’s criterion to establish aspectual
pairs, did not produce coherent results. While some pairs of verbs such as
prygat’-prygnut’ ‘jump’ or videt’-uvidet’ ‘see’ are judged as an aspectual pair by
all who participated, for some pairs such as est’-poest’ ‘eat’, guljat’-poguljat’ ‘go
for a walk’, pet’-propet’ ‘sing’ we see much less consistency. For the last two
pairs in Table 57 three linguists using Maslov’s criterion agree that these verbs
are an aspectual pair, and three other linguists disagree. So as Gorbova (2011:
23) summarizes “even linguists who agree that use of this criterion [Maslov’s
criterion] is reasonable do not get identical results when they analyze the same
material.”27 From this example we can see that Maslov’s criterion is hard to use
��
27 “…даже лингвисты проявляющие единодушие по вопросу о целесообразности
использования данного критерия [критерия Маслова], не получают идентичных
результатов при анализе одного и того же материала” – translated from Russian by J.K.
118 | Constructional profiling and aspectual pairs
and it gives different results when different people use it, which leads us to the
conclusion that its results are unreliable.
The use of syntactic criteria for establishing aspectual pairs are even more ques-
tionable from a cognitive point of view. These criteria make an implicit assump-
tion that in the mind of a native speaker of Russian the imperfective verb is
associated with a perfective verb because perfective and imperfective verbs are
used to describe the same situation in different grammatical environments. So
basically this point of view presupposes that a speaker of Russian hears that (1)
napisal pis’mo ‘writePFV.PAST a letter’ and (2) pišet pis’mo ‘writeIPFV.PRES a letter’ are
used to describe the same situation of writing a letter in the past, however the
former is used when this situation is described in the past tense mode, while the
latter is used when this situation is described in praesens historicum mode.
None of the researchers who formulate criteria for establishing aspectual
pairs mention how the verbs forming a pair are related in the mind of a native
speaker. Is there any specific relation between a perfective verb and an imper-
fective verb from a cognitive point of view? Some scholarly works give us the
impression that their authors do believe that a native speaker somehow sees a
connection between a perfective and an imperfective verb. For example, Mikae-
lian, Shemelev and Zalizniak (2010: 9) when they discuss possible problems
with Maslov’s criterion mention that in the literature it is discussed that “prae-
sens historicum is an artificial context, which a speaker rarely encounters”28.
Mikaelian, Shemelev and Zalizniak reply by saying that “[j]udging by the con-
texts found on the Russian Internet, narrative in praesens historicum is not
peripheral, but a rather widespread form, particularly in colloquial speech” 29
(ibid.). However the whole discussion only makes sense if we assume that
someone (presumably a native speaker of Russian) is establishing what is an
aspectual pair and what is not an aspectual pair based on sentences that he or
she encounters. If the speaker comes across similar sentences in the past tense
and the praesens historicum, then he or she might connect the two verbs found
in those two sentences and consider them to be an aspectual pair. Let us see
��
28 “наст. историческое – это вообще довольно искусственный контекст, с которым
говорящий сталкивается крайне редко” – translated from Russian by J.K.
29 “…[С]удя, например, по текстам, размещенным в русском Интернете, повествование
в наст. историческом – вовсе не периферийная, а очень широко распространенная
форма, в частности, в разговорной речи” – translated from Russian by J.K.
Criteria for establishing an aspectual pair | 119
what experimental studies show about the relation between perfective and
imperfective morphologically related verbs.
Rusakova et al. (2001) and Rusakova and Saj (2003, 2008) study the rela-
tionship between the members of the aspectual pairs in the mind of a native
speaker of Russian using experimental data. Subjects of the study were given a
present tense form of a verb and were asked to produce a past tense form of the
same verb. The formulation of the task slightly, but not significantly, varied
depending on the age group. The same eleven pairs were given to each subject
in the experiment. Subjects of four age groups were tested: three-year-olds,
four-year-olds, five-year-olds and adults as a control group. Each participant
received as a stimulus only one element from each pair. Some participants were
given the imperfective stimulus and some – the perfective stimulus. The partici-
pants from all groups produced as a response not only the form from the same
paradigm as the stimulus, but also the forms from the paradigm of its aspectual
counterpart. For example, the subjects were given the form kidaet ‘throwIPFV.PRES’.
96% of the participants produced the response from the same paradigm kidal
‘throwIPFV.PST’, while 4% of the participants produced the perfective past tense
form kinul ‘throwPFV.PST’. Even more paradigm deviation was observed for the
perfective stimuli: the subjects given the perfective present tense form kinet
‘throwPFV.PRES’ in 43% of all responses gave the perfective past tense form and in
57% of the responses gave the imperfective past tense form. Overall paradigm
misplacement was observed in 4-36% for imperfective stimuli and in 40-77% for
perfective stimuli. Rusakova and Saj argue that this difference between the
behavior of the imperfective and perfective stimuli reflects the fact that imper-
fective verb forms are more basic in the mind of a Russian speaker. Even more
interestingly three-year-olds in Rusakova and Saj’s experiment show more par-
adigm awareness than all older children and adults, see results distributed by
age groups in Table 58. The difference between the results for the three-years-
olds and adults is statistically significant.
age 3 4 5 adults
participants in the group 20 25 21 75
correct answers 73% 67% 61% 61%
Rusakova and Saj (2008) point out that their observations are in agreement with
two well established notions: 1) Russian speaking children even in early stages
of language development are able to choose an appropriate aspect form (see
Cejtlin 2000: 89, 148-152); 2) at a certain age Russian speaking children produce
120 | Constructional profiling and aspectual pairs
errors in verb forms which indicate that they learn how to derive one aspectual
form from the other (see Cejtlin 2009: 7). So it seems that Russian speaking
children first acquire a full verbal paradigm of each verb and master the use of
aspect in a given sentence and only later produce a more complicated paradigm
that unites both imperfective and perfective forms. Rusakova and Saj conclude
that the members of aspectual pairs are not independent in the mental lexicon
of a native speaker of Russian. They are united into an aspectual hyperpara-
digm. The aspectual relations are used by speakers in the process of speech,
which can be observed in experimental conditions. Relations between the
members of aspectual pairs are different for different pairs.
Rusakova and Saj’s experiment shows that in the mind of a speaker there is
a relation between imperfective and perfective verbs. However in this experi-
ment the eleven pairs chosen for the study present an unusual sample of the
Russian aspectual pairs in terms of derivation. For ten of the verbs the imperfec-
tive verb is produced from the perfective by adding a suffix. For four prefixed
verbs this is a standard derivation within an aspectual pair (vydumyvat’-
vydumat’ ‘invent’, oskorbljat’-oskorbit’ ‘insult’, obmanyvat’-obmanut’ ‘deceive’,
srezat’-srezat’ ‘cut off’). These are typical suffixal aspectual pairs, which most
scholars consider to be “true” aspectual pairs. The majority of aspectual pairs in
Russian are formed via this mechanism. The other six verbs present an unusual
case where the base is a bare perfective (kidat’-kinut’ ‘throw’, davat’-dat’ ‘give’,
brosat’-brosit’ ‘cast’, pomogat’-pomoč’ ‘help’, puskat’-pustit’ ‘let’, rešat’-rešit’
‘decide’). These pairs present a very small and closed subclass of Russian aspec-
tual pairs. Finally only one of the tested pairs was derived from an imperfective
base via prefixation (bežat’-probežat’ ‘run’30). This underrepresents the for-
mation of aspectual pairs via prefixation, which is the second most frequent
way to produce an aspectual pair and which is the focus of this chapter.
It is most probable that derivational relations between the members of a
pair affect how the members of a pair are stored in a speaker’s mind. It would be
interesting to explore what responses native speakers of Russian would give in
a similar experiment including more prefixed pairs. The research done by the
CLEAR group at the Arctic University of Norway (see Janda et al. 2013) suggests
that a speaker of Russian would recognize several prefixed verbs as possible
aspectual counterparts of one non-prefixed imperfective verb. Different perfec-
tive counterparts would be chosen based on context and construction in which
||
30 It has to be noted that not all sources confirm that these two verbs form a pair. For example,
the “Exploring Emptiness” database (emptyprefix.uit.no) based on several dictionaries of
Russian does not contain this pair.
Criteria for establishing an aspectual pair | 121
an imperfective is used. Thus it seems that even though perfective and imper-
fective verbs are cognitively related, this connection is usually more complicat-
ed than a simple one-to-one correspondence. It is more likely that a native
speaker of Russian operates with a network of perfectives and imperfectives
related to each other. The choice of a concrete form is made based on context
and the construction in which the verb is used.
It is important to note here that all the criteria for establishing aspectual pairs
are based on a very specific view of what a verb is. The verb is seen as a mono-
lithic configuration that can only have one possible argument structure.
“[M]any theories of grammar have been built on the assumption that the syntac-
tic realization of arguments – their category type and their grammatical func-
tion – is largely predictable from the meaning of their verbs” as Levin and Rap-
poport (2005: 7) point out. Claims have often been made that a verb
unambiguously selects arguments bearing its semantic roles, e.g. the verb hit
selects an agent and a patient (Chomsky 1986: 86). Similar structures are pro-
posed for eat (Pesetsky 1995: 4) and see (Rothstein 1983: 23). However, if we
turn to authentic usage, we find that a verb is usually scattered among several
possible argument structures. For example, these accounts do not consider the
fact that it is possible for the verb eat to omit the patient with indefinite inter-
pretation; see (61), while no such contexts are possible for the verb hit. The verb
see also may omit the object, however in this case it will receive a definite inter-
pretation as in (62), see (Fillmore 1986).
(61) The group around me stared as I ate. (Nathan Hoturoa Gray. 2008. Wall
of Wonder. National Geographic. Vol. 7, Iss. 7)31
(62) What are you talking about? I was there. I saw. (Michael Libling. 2010. If
You've Ever Been a Lady. Fantasy & Science Fiction. Vol. 111, Iss. 3; pg.
141)
As Levin and Rappaport (2005: 9) observe, “[s]uch examples show that the pro-
gram of deriving a verb’s argument realization options from its meaning must
||
31 This example and English examples below are taken form the COCA corpus:
http://corpus.byu.edu/coca/
122 | Constructional profiling and aspectual pairs
be firmly grounded in the relevant empirical facts, and the range of facts goes
well beyond those that syntacticians usually cite.” Moreover even more seman-
tically straightforward verbs such as kick allow at least eight distinct argument
structures (Goldberg 1995: 11):
tures. Acknowledgement of this fact transforms our view of aspectual pairs and
the use of syntactic criteria for establishing them.
How can we reformulate the criteria in terms of constructions? Let us as-
sume that we have two morphologically related verbs which are candidates for
an aspectual pair. Aspectual interchangeability means that we have a construc-
tion that allows both perfective and imperfective verbs. This construction can be
used in several grammatical contexts: past tense, praesens historicum, habitu-
al, imperative with and without negation. What this construction describes can
be seen as the same situation ignoring the differences attributable to grammati-
cal contexts. Therefore we can discover parallelism between the perfective and
imperfective verb used in the two variants of the construction.
It is essential to add here that in defining parallelism in terms of construc-
tions we safely avoid the problems of universal vs. existential formulation of the
criteria. Now the parallel is drawn not between the two verbs, but between the
two uses of a construction. Now we can reformulate the hypothesis about two
verbs in terms of constructions. The aspectual relation between the two verbs
can be stated in terms of how many constructions allow both verbs. Thus, we
approximate how frequently the simplex imperfective verbs and their perfec-
tives with the prefix pro- serve as aspectual counterparts of each other.
The advantage of using this measure is the fact that it is based on a corpus
sample, so it gives us a usage-based metric of the interchangeability of the two
verbs. Therefore our measurements model how well the two verbs function as
an established aspectual pair in the mind of a native speaker of Russian32.
Among pairs there are more prototypical and less prototypical aspectual pairs.
These methods allow us to measure numerically the prototypicality of a given
pair. In addition I show that prototypicality correlates with how well the mean-
ing of the base imperfective intersects with the meaning of the prefix pro-. This
finding supports the Overlap hypothesis.
The Overlap hypothesis concerns perfective verbs derived from simplex
imperfective verbs via prefixation. Since the perfective and the imperfective
share the same lexical meaning, according to some scholars prefixes in such
cases are considered to be meaningless. The Overlap hypothesis, however,
states that the prefix has meaning but its meaning overlaps with the meaning of
the simplex verb. The Overlap hypothesis was first introduced in van Schoone-
veld 1958 and Isačenko 1960 (cf. also Vey 1952 with reference to Czech) and has
||
32 The measurement would be even more accurate if I used the spoken corpus of modern
Russian. However, since a large spoken corpus of Russian is unavailable, the analysis based on
the written corpus of Russian gives us the best achievable approximation.
124 | Constructional profiling and aspectual pairs
recently been argued for in Janda et al. 2013. The overlap is a scalar phenome-
non, which means that it can be more or less complete. This chapter shows that
the interchangeability of a simplex imperfective and a prefixed perfective de-
pends on how well the meaning of the simplex verb overlaps with the meaning
of the prefix. This is a strong argument in favor of the Overlap hypothesis. This
correlation would not be present if the meaning of the prefix in such perfectives
was empty.
properly (here stages needed to reach the fully cooked phase function as a
quantum through which we travel). Thus, the prefix pro- is an ideal candidate
for a prefix with consistent semantics and with more than 100 aspectual pairs
formed via prefixation.
Intersecting prefixed aspectual pairs formed via the prefix pro-, verbs that
have at least 100 examples in the RNC, and pairs where either the imperfective
or the perfective are mentioned in the Apresjan and Pall’s dictionary, we receive
seventeen alleged pairs. These are the verbs investigated in this study. If we
look at these seventeen pairs, we see that among the pairs there are many
speech and sound verbs. Janda and Lyashevskaya (2013) study the semantic
classes of the verbs attracted and repulsed from the prefix pro-. Janda and
Lyashevskaya adopt the semantic classes tagged in the RNC and based on that
show that verbs of speech and sound are particularly attracted to the prefix pro-.
This attraction occurs because the meaning of the prefix pro- is often related to
“penetration of sound waves through space” (ibid.: 21). From the point of view
of the semantic classes tagged in the RNC, among the seventeen imperfectives
there are five verbs of sound (gremet’ ‘thunder’, gudet’ ‘buzz’, zvenet’ ring,
clank’, zvučat’ ‘sound’, and pet’ ‘sing’), four speech verbs (informirovat’ ‘in-
form’, kričat’ ‘shout’, citirovat’ ‘cite’, and šeptat’ ‘whisper’), two verbs of move-
ment (vesti ‘lead’ and tjanut’ ‘pull’), one mental verb (čitat’ ‘read; recite’), one
verb of impact (bit’ ‘beat’), one verb of interaction (kontrolirovat’ ‘control’), one
verb of perception (demonstrirovat’ ‘demonstrate’) and two untagged verbs
(golosovat’ ‘vote; hitch a ride’ and dlit’sja ‘last’).
The 3400 (17x2x100) examples investigated in this study are tagged using
the FrameBank tagging device. FrameBank (www.framebank.ru) is an annota-
tion project that takes as input examples from the Russian National Corpus. The
goal of the project is to mark the set of arguments and adjuncts of a given predi-
cate. The project is based on Construction Grammar theory and continues the
tradition of lexical resources such as FrameNet (Johnson et al. 2003, VerbNet
(Kipper et al. 2006), PropBank (Palmer et al. 2005), NomBank (Meyers 2007),
and Vallex (Lopatková et al. 2006) developed for English, Czech and other lan-
guages. Within the FrameBank each sentence is tagged with a construction
realized in that sentence.
All seventeen imperfective verbs in this study are described in Apresjan and
Pall’s dictionary. Three perfective verbs probit’ ‘beatPFV’, provesti ‘leadPFV’ and
protjanut’ ‘pullPFV’ are also described in the dictionary. The constructions for the
remaining verbs are ascribed by means of the FrameBank system, following the
main ideas and methodology of Apresjan and Pall’s study. If a verb described in
the dictionary is attested in a construction not ascribed to it, a new construction
Methodology of the study | 127
(71) Citiruja puškinskie stroki, S. Vavilov otmečaet, čto ètot otryvok «genialen
po svoej glubine i značeniju dlja učenogo», ibo «svidetel’stvuet o
proniknovennom ponimanii Puškinym metodov naučnogo tvorčestva».
[A. K. Suxotin. Paradoksy nauki (1978)]
‘Quoting Pushkin's lines, S. Vavilov notes that this passage is "brilliant
in its depth and meaning for a scientist," because it is "the evidence for
Pushkin’s innate understanding of methods of scientific creativity.”’
Table 59 presents all seventeen pairs investigated in this study and shows the
number of constructions found for imperfective and perfective verbs. Pairs are
given in Latin alphabetic order for the imperfectives. Thus, we see, for example,
that the imperfective bit’ ‘beatIPFV’ is found in 38 constructions, whereas the
perfective probit’ ‘beatPFV’ is found in 21 constructions. Table 59 shows us that
overall in a pair imperfective verbs tend to have more constructions than perfec-
tive verbs: this is true for thirteen out of seventeen studied pairs.
128 | Constructional profiling and aspectual pairs
As we can see from the number of constructions attested for each verb, no pair
of verbs shows truly unified behavior. The constructional profiles of the verbs
contain between four (for the verbs kontrolirovat’ ‘control’ and its alleged per-
fective counterpart prokontrolirovat’) and forty-three (for the verb tjanut’ ‘pul-
lIPFV’) constructions. No verb is found in one and only one construction, moreo-
ver no verb is found in only two or three constructions. For illustrative
purposes, the constructional profiles of the verbs kontrolirovat’ ‘controlIPFV’ and
tjanut’ ‘pullIPFV’ are shown on the charts in Figure 18 and Figure 19. Figure 18
shows us that the verb kontrolirovat’ ‘controlIPFV’ is used most frequently in the
transitive construction NPnom V NPacc (eighty-nine out of 100 examples). The
second construction in terms of frequency construction, the intransitive NPnom
V, appears notably less frequently and is attested in six examples. The remain-
ing two constructions are even less frequent: one of them, NPnom V NPacc s +
pomošč'ju + NPgen, refers to control via something and is found in three exam-
ples and the other, the embedded clause construction NPnom V Conj + CL, is
found in only one example.
Methodology of the study | 129
Fig. 18: The constructional profile of the verb kontrolirovat’ ‘control’ (4 constructions)
The opposite extreme is presented by the verb tjanut’ ‘pullIPFV’ which is found in
forty-three different constructions. Each construction found for the verb tjanut’
‘pullIPFV’ is attested for a small number of sentences. Because of that the scale for
Figure 19 is smaller than for Figure 18. While the maximum on Figure 19 is 100
examples, the maximum on Figure 19 is only twelve, otherwise the number of
attested sentences would not be distinguishable on the chart. We can see that
the most frequent construction NPnom V {ADV/PRto NPx} for tjanut’ ‘pullIPFV’ is
attested in eleven examples. The next construction in terms of frequency, NPacc
Vimpers {ADV/PRto NPx}, is attested in eight examples. There are two construc-
tions found in five examples, four constructions found in four examples each,
five constructions found in three examples each, nine constructions attested
only in two examples, and twenty-one constructions found in only one exam-
ple. We see that the verb tjanut’ ‘pullIPFV’ has multiple constructions without any
clear central construction.
130 | Constructional profiling and aspectual pairs
Fig. 19: The constructional profile of the verb tjanut’ ‘pull’ (43 constructions)
However a typical verb in the sample lies somewhere between the most unified
verbs kontrolirovat’ and prokontrolirovat’ ‘control’ and the most diverse verb
tjanut’ ‘pullIPFV’. The verb kričat’ ‘shoutIPFV’ may be viewed as an example of a
typical verb in the given sample. We see in Figure 20 that this verb has prefer-
ence for two constructions: the direct speech construction NPnom V “CL” (sixty
examples) and the intransitive construction NPnom V (twenty-three examples),
while another ten constructions are notably less frequent and only appear in
one, two or three examples.
Methodology of the study | 131
Fig. 20: The constructional profile of the verb kričat’ ‘shout’ (12 constructions)
It seems reasonable to include in the verb’s constructional profile only the con-
structions that appear frequently among the examples of a verb. I pose a
threshold of five attested examples for including a construction in a construc-
tional profile. This means that from now on the term “constructional profile”
refers to the set of constructions that are found in five or more examples among
the randomly selected one hundred examples of a verb. Constructions that are
found in at least five examples are referred to as “frequent constructions.”
For example, the constructional profile of the verb kričat’ ‘shoutIPFV’ con-
tains only two frequent constructions (the direct speech construction NPnom V
“CL” and the intransitive construction NPnom V), the constructional profile of
the verb kontrolirovat’ ‘conrolIPFV’ also consists of two frequent constructions
(NPnom V NPacc and NPnom V), while the constructional profile of the verb
tjanut’ ‘pullIPFV’ contains four frequent constructions (NPnom V {ADV/PRto NPx},
NPacc Vimpers {ADV/PRto NPx}, NPacc Vimpers Vinf, NPnom V {ADV/PRtraject
NPx}). See the constructional profiles of the alleged aspectual partners kontroli-
rovat’ and prokontrolirovat’ ‘control’ after the application of the threshold in
Tables 60 and 61.
132 | Constructional profiling and aspectual pairs
Table 60: Constructional profile of the verb kontrolirovat’ ‘controlIPFV’ (4 constructions, 2 fre-
quent constructions)
If we want to find out how compatible the constructional profiles of the imper-
fective and the perfective are, we need to find the intersection of their construc-
tional profiles. Let us imagine the constructional profiles of the two verbs as
molecular graphs, as in Figure 21. The intersection of the two profiles can be
seen if we superimpose one graph on the other as in Figure 22. Here the striped
area shows the intersection of the two constructional profiles. We can see that
for the transitive construction the intersected area is smaller than the area for
the imperfective verb kontrolirovat’ ‘controlIPFV’, while for the intransitive con-
struction NPnom V the intersected area is smaller than the area of the same
construction for the perfective verb prokontrolirovat’ ‘controlPFV’. Let us find a
number that characterizes the striped intersection area in Figure 22.
Methodology of the study | 133
Fig. 21: Graphs of constructional profiles of the verbs kontrolirovat’ and prokontrolirovat’
‘control’
Fig. 22: Superimposed graphs of constructional profiles of the verbs kontrolirovat’ and pro-
kontrolirovat’ ‘control’
134 | Constructional profiling and aspectual pairs
Table 62 shows the two constructions that are attested in the constructional
profiles of both the imperfective verb kontrolirovat’ ‘controlIPFV’ and the perfec-
tive verb prokontrolirovat’ ‘controlPFV’. The second and third columns of the
table present the number of examples of each construction attested for each
verb. The last column shows the minimum of the two numbers presented in
each row. For the construction NPnom V NPacc the minimum of attested exam-
ples is eighty – the number of examples of this construction for the verb pro-
kontrolirovat’ ‘controlPFV’; for the construction NPnom V the minimum of attest-
ed examples is six – the number of examples of this construction for the verb
kontrolirovat’ ‘controlIPFV’. Thus, the intersection rate for this pair of verbs is 80 +
6 = 86.
Table 62: Intersection of constructional profiles of the verbs kontrolirovat’ and prokontroliro-
vat’ ‘control’
minimum attested
construction kontrolirovat’ prokontrolirovat’ examples
NPnom V NPacc 89 80 80
NPnom V 6 7 6
Intersection rate 86
All seventeen pairs of verbs explored in this chapter and their intersection rates
are presented in Table 63. The verbs are ordered according to the intersection
rate from highest to lowest. The second column of Table 63 shows us that no
pair has the same constructional profiles and the same distribution within the
profiles. If a pair were to have a truly unified behavior, both the imperfective
and the perfective would be found in only one construction and the intersection
rate would be 100. As we can see from Table 63 this is not the case for even most
unified verbs in the studied sample. The maximum intersection rate is 86, which
means that even the best pair in the sample shows some differences in terms of
constructional profiles.
The intersection rate is correlated with the productivity of the grammatical
class of the verb: verbs with higher intersection rates belong to productive
grammatical classes, while verbs with lower intersection rates belong to non-
productive grammatical classes. The grammatical classes of the studied verbs
are analyzed using two sources: Zalizniak’s grammatical dictionary (1980) and
Townsend’s classification of Russian paradigms (1968: 97-114). The third and
fourth columns of Table 63 contain the index and type frequency of the gram-
matical class from Zalizniak’s dictionary. The fifth and sixth columns of Table
63 show the grammatical class and productivity of that class according to
Townsend’s classification. Grammatical classes are shown for the imperfective
verb in a pair33.
Let us consider the verbs with higher intersection rates. From the pair
kontrolirovat’/prokontrolirovat’ ‘control’ to the pair citirovat’/procitirovat’ ‘cite’
which have intersection rates between 86 and 67, the grammatical classes are
listed as productive in Townsend’s classification and are attested more than
2000 times in Zalizniak’s dictionary. Here we have pairs of verbs marked with
the suffix -ova- which in modern Russian is associated with loan words:
kontrolirovat’ ‘controlIPFV’, demonstrirovat’ ‘demonstrateIPFV’, informirovat’ ‘in-
fromIPFV’ and citirovat’ ‘citeIPFV’. All these verbs belong to class 2a in Zalizniak’s
classification or the OVA-type in Townsend’s classification. We see that there
||
33 For every pair except the pair kričat’/prokričat’ ‘shout’ the grammatical class for the perfec-
tive verb matches the grammatical class of the imperfective in both sources. The grammatical
class of the perfective prokričat’ in Zalizniak’s dictionary is 5b## and this class is attested for
thirty-seven verbs. According to Townsend’s classification verbs kričat’ and prokričat’ belong
to the same ŽA-class.
136 | Constructional profiling and aspectual pairs
are 2928 verbs of this class found in Zalizniak’s dictionary. In addition we have
two pairs of verbs dlit’sja/prodlit’sja ‘last’ and čitat’/pročitat’ ‘read; recite’. The
first one belongs to the 4b class in Zalizniak’s classification and the I-type in
Townsend’s classification. The second verb belongs to class 1a according to
Zalizniak and the AJ-type according to Townsend. Class 4b is attested for 2605
verbs and class 1a is attested for 12432 verbs.
productive
‘demonstrate’ 75 2a 2928 OVA
informirovat’/proinformirovat’ ‘in-
form’ 72 2a 2928 OVA
čitat’/pročitat’ ‘read; recite’ 68 1a 12432 AJ
golosovat’/progolosovat’ ‘vote; hitch
a ride’ 67 2a 2928 OVA
citirovat’/procitirovat’ ‘cite’ 67 2a 2928 OVA
kričat’/prokričat’ ‘shout’ 65 5b 387 ŽA
šeptat’/prošeptat’ ‘whisper’ 60 6c 510 A
zvučat’/prozvučat’ ‘sound’ 55 5b 387 ŽA non-productive
gremet’/progremet’ ‘thunder’ 51 5b 387 E
zvenet’/prozvenet’ ‘ring, clank’ 50 5b 387 E
vesti/provesti ‘lead’ 33 7b/b 126 D
pet’/propet’ ‘sing’ 23 12b 6 irregular
gudet’/progudet’ ‘buzz’ 15 5b 387 E
tjanut’/protjanut’ ‘pull’ 0 3c 47 NU
bit’/probit’ ‘beat’ 0 11b 33 IJ
All verbs in Table 63 starting from the pair kričat’/prokričat’ ‘shout’ (intersection
rate 65 and lower) belong to closed non-productive classes according to Town-
send and their classes are attested for less than 600 verbs in Zalizniak’s diction-
ary. Here we see the verbs kričat’ ‘shout’, zvučat’ ‘sound’, gremet’ ‘thunder’,
zvenet’ ‘ring, clank’, and gudet’ ‘buzz’ classified as 5b in Zalizniak’s dictionary
(attested for 387 verbs). The first two of these belong to the ŽA-type and the last
three to the E-type according to Townsend. In addition to that we see the pair
šeptat’/prošeptat’ ‘whisper’, which belongs to Townsend’s A-type and Zalizni-
Intersection rates for seventeen pairs of verbs | 137
ak’s grammatical class 6c with 509 other verbs. The remaining four verbs all
belong to unique or even irregular grammatical classes attested in less than 200
verbs each, and require specific discussion.
The paradigms of these four verbs contain alternations that are irregular
from the point of view of modern Russian and are the results of historical pho-
nological changes in Common Slavic or Old Russian (see, for example,
Trubetzkoy 2001: 92, Townsend 1968: 98-111). The verb pet’ ‘singIPFV’ has the
present tense form poet ‘singPRES3SG’. The verb bit’ ‘beatIPFV’ has the present tense
form b’et ‘beatPRES3SG’. The verb vesti ‘leadIPFV’ has the present tense form vedet
‘leadPRES3SG’ and forms prefixed secondary imperfectives using the stem -vodit’
(privodit’ ‘bring’). The verb tjanut’ ‘pull’ in other instantiations of the root has
-g- (tjagat’ ‘pull heavy weight (coll.)’, zatjagivat’ ‘tighten’) that disappeared in
front of the suffix -nu-. Thus, these four verbs can be viewed as the most native
verbs in the studied sample. We would expect the native words to have a more
widely dispersed network of meanings and therefore to be attested in more
constructions.
On the contrary, we would expect the above mentioned borrowed verbs
with the -ova- suffix to have more focused semantics. More focused semantics
entails that fewer constructions are available for a verb, and a smaller number
of possible constructions increases the probability that the constructional pro-
files of the imperfective and perfective verbs will share constructions. The result
of this effect would be a higher intersection rate for the loan words and a lower
intersection rate for the native words. As we have seen the four loan verbs,
kontrolirovat’ ‘controlIPFV’, demonstrirovat’ ‘demonstrateIPFV’, informirovat’ ‘in-
fromIPFV’ and citirovat’ ‘citeIPFV’, have high intersection rates (86, 75, 72 and 67),
while the four native verbs, vesti/provesti ‘lead’, pet’/propet’ ‘sing’, tja-
nut’/protjanut’ ‘pull’ and bit’/probit’ ‘beat’, have low intersection rates (33,
which in section 5.3.4 is argued to be 0, 23, 0 and 0). Thus, we see that inter-
changeability between imperfective and perfective verbs is affected by how long
the word has been present in the language and the size of the network of sub-
meanings the word has produced over this time. Correlation of productivity with
intersection rate can be seen as an extension of this effect. Verbs in productive
classes tend to have fewer available constructions and less diversity, which
correlates with higher intersection rate. Non-productive verbs are usually those
that have been in the language for a long time, and as a result are more dis-
persed in their meaning and participate in more possible constructions, which
correlates with lower intersection rate.
Appendix contains the constructional profiles of all investigated verbs and
tables showing how the intersection rate is calculated for every pair. Here in the
chapter I discuss the main trends and show how the meaning of the prefix pro-
138 | Constructional profiling and aspectual pairs
affects the constructional profiles of the prefixed perfectives. The verbs will be
discussed in four subsections. The first subsection (5.3.1) will present the four
pairs of verbs with the highest intersection rate: kontrolirovat’/prokontrolirovat’
‘control’, dlit’sja/prodlit’sja ‘last’, demonstrirovat’/prodemonstrirovat’ ‘demon-
strate’, golosovat’/progolosovat’ ‘vote; hitch a ride’. The second subsection
(5.3.2) explores the pairs of speech and mental verbs. The list contains four pairs
of speech verbs (informirovat’/proinformirovat’ ‘inform’, citirovat’/procitirovat’
‘cite’, kričat’/prokričat’ ‘shout’ and šeptat’/prošeptat’ ‘whisper’) and one pair of
mental verbs čitat’/pročitat’ ‘read; recite’. The third subsection (5.3.3) describes
the behavior of five pairs of sound verbs: zvučat’/prozvučat’ ‘sound’,
gremet’/progremet’ ‘thunder’, zvenet’/prozvenet’ ‘ring, clank’, pet’/propet’ ‘sing’
and gudet’/progudet’ ‘buzz’. The fourth subsection (5.3.4) discusses the three
verb pairs with the lowest intersection rate: vesti/provesti ‘lead’, bit’/probit’
‘beat’ and tjanut’/protjanut’ ‘pull’. If we compare these subsections to the order-
ing in Table 63, we will see that these four classes correspond fairly well to the
order of intersection rates, with a few deviations. Among the speech and mental
verbs we find the pair golosovat’/progolosovat’ ‘vote; hitch a ride’, which is
discussed among the verbs with the highest intersection rate. Among the verbs
of sound we find the pair vesti/provesti ‘lead’ which is discussed together with
the pairs vesti/provesti ‘lead’ and bit’/probit’ ‘beat’ – the verbs with the lowest
intersection rate. Both the semantic groupings and the scale of intersection
rates provide interesting insights into the nature of intersection, so I group the
verbs as described above. The four subsections described above are followed by
a discussion of the problem of prefix variation as related to the pairs in this
study (5.3.5).
The verbs dlit’sja and prodlit’sja ‘last’ both strongly coincide with the semantics
of the prefix pro-. This prefix often means spending an amount of time being
busy with an activity. This can be seen from the fact that the prefix pro- often
combines with accusative NPs denoting time as in (73)-(75).
(73) My proboltali počti tri časa… [Ja želanna. Razve èto stydno? // «Daša»,
2004]
‘We chatted for almost three hours...’
(74) Rebenok, po krajnej mere, prospal vse poltora časa ot doma do doma.
[Andrej Kolesnikov. Bubliki Mondeo (2002) // «Avtopilot», 2002.01.15]
‘The child, at least, slept an hour and a half from door to door.’
(75) Tol’ko odin raz zabrela v magazin igrušek i vmesto urokov prostojala
četyre časa u vitrin. [Natal’ja Skljarova. Kazaki-razbojniki (2002) //
«Večernjaja Moskva», 2002.01.10]
‘Only once she wandered into a toy store and instead of school she
stood for four hours in front of shop windows.’
Thus, it is not surprizing that the verbs dlit’sja and prodlit’sja ‘last’ have a high
intersection rate, 79, the second highest intersection rate among the seventeen
explored pairs. The large overlap between the meaning of the prefix pro- and
the semantics of the imperfective verb dlit’sja ‘lastIPFV’ make the perfective verb
prodlit’sja ‘lastPFV’ very similar to the imperfective verb. The only difference
between the two verbs is that the perfective verb focuses more on the end of the
time period. The only construction which appears in the constructional profile
of the imperfective verb dlit’sja ‘last ’and is not present in the profile of the
perfective verb prodlit’sja ‘lastPFV’ is the construction that marks both the
beginning and the end of the time period NPnom V {ot + NPgen / s + NPgen} do +
140 | Constructional profiling and aspectual pairs
NPgen; see (76). The contexts that mark the beginning of a time period cannot
be described with the perfective prodlit’sja ‘lastPFV’. At the same time the con-
struction that marks only the end of a time period NPnom V do + NPgen is more
frequent among the uses of the perfective verb. The verb dlit’sja is attested in
seven examples of this construction, while the verb prodlit’sja is found in twen-
ty examples of it. Focus on the end of a time period seems natural for the perfec-
tive, which stresses the resulting stage of the event.
||
34 This example, and other examples in this chapter for which the source is not specified, are
from the FrameBank database and AbbyLingvo.Pro Russian-English dictionary (see,
http://lingvopro.abbyyonline.com/en/).
Intersection rates for seventeen pairs of verbs | 141
Golosovat’ and progolosovat’ ‘vote’ share the same profiles consisting of three
constructions: the intransitive construction NPnom V, the construction with the
preposition za ‘for’ NPnom V za + NPacc, and the construction with the preposi-
tion protiv ‘against’ NPnom V protiv + NPgen. However the two verbs have dif-
ferent foci within their similar profiles. The perfective verb has stronger prefer-
ence for the ‘for’-construction than the imperfective. The imperfective verb
golosovat’ ‘voteIPFV’ has forty-three uses of the ‘for’-construction, while the per-
fective progolosovat’ ‘votePFV’ has sixty-three uses of the ‘for’-construction; see
(79) and (80). This asymmetry seems natural, because when we refer to a vote
that has already happened we are more likely to be interested in the votes for
something than the votes against something.
Summing up, the four verbs described in this section have tightly focused
meanings and have few submeanings. This entails that uses of a verb are not
diversely spread among several constructions, but mostly belong to one con-
struction, which increases the possibility of a high intersection rate. The mean-
ings of all four imperfective verbs are characterized by large overlap with the
meaning of the prefix pro-. Therefore the meanings of the perfectives created by
adding the prefix pro- to the imperfective verbs are very similar to the meanings
of the original imperfectives. Thus a high intersection rate is associated with the
amount of overlap between the meaning of the imperfective and the meaning of
the prefix pro-.
142 | Constructional profiling and aspectual pairs
This section explores five pairs of verbs: four pairs of speech verbs and one pair
of mental verbs čitat’/pročitat’ ‘read; recite’; see Table 65.
Table 65: The intersection rates of the pairs of speech and mental verbs
We recall that verbs of speech and sound are expected to be found among the
aspectual pairs formed with the prefix pro-. Janda and Lyashevskaya (2013)
show that verbs of speech and sound exhibit the strongest attraction for the
prefix pro- among all the semantic classes and five prefixes they study in the
paper. The four pairs of speech verbs explored in this study produce a coherent
cluster with an intersection rate between 72 and 60. Comparing profiles of these
pairs we can see several patterns.
Let us start with the pair informirovat’/proinformirovat’ ‘inform’. The imper-
fective verb has more uses of the descriptive construction with the preposition o
‘about’ NPnom V NPacc o + NPloc (sixty-five examples as opposed to fifty-three);
see (81) and (82).
The other three constructions, the transitive construction NPnom V NPacc, the
direct speech construction NPnom V NPacc "CL" and the embedded clause con-
struction NPnom V NPacc Conj + CL, have twice as many uses of the perfective
compared to the imperfective (seventeen as opposed to eight, thirteen as op-
posed to six, and eleven as opposed to five respectively). Preference of perfec-
Intersection rates for seventeen pairs of verbs | 143
The verb prokričat’ ‘shoutPFV’ on the other hand shows a preference for the tran-
sitive construction NPnom V NPacc: this construction only appears in the con-
structional profile of the perfective verb and is absent from the constructional
144 | Constructional profiling and aspectual pairs
The verbs šeptat’ ‘whisperIPFV’ and prošeptat’ ‘whisperPFV’ are found in very dif-
ferent sets of constructions. The imperfective verb šeptat’ ‘whisperIPFV’ is attest-
ed in five frequent constructions, and the perfective verb prošeptat’ ‘whisperPFV’
is attested in three frequent constructions, however the verbs share only two of
those constructions: the direct speech construction NPnom V «CL» and the
ditransitive construction NPnom V {NPacc / Conj + CL / CL} NPdat. As a result
the alleged aspectual pair has the intersection rate 60. In addition to the shared
constructions, the perfective verb is also found in a construction with a lexical-
ized phrase referring to the addressee’s ear NPnom V [{NPacc / «CL»}] [NPdat]
{na + uxo / v + uxo} ‘whisper in one’s ear’; see (87). This construction appears
among the uses of the imperfective verb šeptat’ ‘whisperIPFV’, but is attested only
in three examples, and therefore this construction does not appear in the con-
structional profile of the verb. The imperfective verb, in addition to the two
shared constructions, is also found in a construction with the transitive NPnom
V NPacc, the intransitive construction NPnom V, and the reciprocal construction
in which the verb has the reciprocal marker -sja- NPnom Vsja [s + NPins] [{о +
NPloc / "CL" / CL}]; see (88).
(87) Gusev šagnul k nemu vplotnuju i prošeptal v samoe uxo: “A pered toboj
dolžok.”
‘Gusev stepped close to him and whispered in his ear: “You owe me.”’
Intersection rates for seventeen pairs of verbs | 145
The verbs šeptat’ ‘whisperIPFV’ and prošeptat’ ‘whisperPFV’, which have the lowest
intersection rate among the speech and mental verbs investigated in this study,
represent a clear case of different foci within the constructional profiles. While
the direct speech construction NPnom V «CL» is attested in approximately one
half of the examples of the imperfective verb (fifty-five examples), it is more
dominant for the perfective verb, where it is found in more than 4/5 of all uses
(eighty-three examples). We will see similar differences playing an even more
important role when we turn to the only remaining verb pair in this section.
Last but not least, let us discuss the only pair of mental verbs in this class:
čitat’ and pročitat’ ‘read; recite’. This pair is united with the verbs of speech,
because Russian verbs čitat’ and pročitat’ combine two different, but related
meanings: ‘read’ referring to processing of a written text silently and ‘recite’
denoting reciting a written text aloud. While the first meaning is the most prom-
inent and frequent, and it is due to that meaning that this verb is classified as a
mental verb, the second meaning also occurs among the uses of both the imper-
fective and the perfective verbs and plays an important role in the construction-
al profiles of both verbs. Constructions of reciting relate the verbs čitat’ and -
pročitat’ ‘read; recite’ to the verbs of speech. Both verbs when used to denote
‘recite’ are found in the transitive construction NPnom V NPacc where the direct
object usually refers to a recited text. This construction is attested in thirteen
examples of the imperfective verb čitat’ ‘read; recitePFV’ and nine uses of the
perfective verb pročitat’ ‘read; recitePFV’. Thus the submeaning ‘recite’ contrib-
utes nine examples of shared construction to the intersection rate of the two
verbs. Only one other construction is shared by both verbs: the construction
with the direct object with the submeaning ‘read’ NPnom V NPacc (fifty-nine
examples of čitat’ and sixty-six examples of pročitat’). In addition to this the
imperfective verb has a preference for the submeaning ‘read’ where it is attested
in the intransitive construction NPnom V and the embedded clause construction
NPnom V {Conj + CL / CL} (see (89) and (90)), while the perfective verb is used in
one additional construction distinctive for the submeaning ‘recite’. This is the
direct speech construction NPnom V «CL» (see (91)), which again as for the verb
prošeptat’ ‘whisperPFV’ shows us that the perfective verbs with the prefix pro- are
more likely to be used as verbs of speech than their imperfective counterparts.
This tendency is attributed to the meaning introduced by the prefix pro-.
Summing up, verbs of speech, plus the mental verbs čitat’ and pročitat’ ‘read;
recite’ which are semantically and syntactically contiguous to them, all have
high intersection rates. We see that the meaning of a speech verb overlaps with
the meaning of the prefix pro- which refers to the sound traveling through space
and as a result the imperfective and the perfective with the prefix pro- show
much similarity in their constructional profiles. As we proceed to the verbs that
have smaller semantic overlap with the meaning of the prefix pro-, we will see
that the effect of the semantics of the prefix pro- will become more and more
prominent.
Sound verbs overall have lower intersection rates than the verbs of speech; see
Table 66. Among verbs of sound we can see two smaller clusters. The first three
pairs of verbs zvučat’/prozvučat’ ‘sound’, gremet’/progremet’ ‘thunder’ and
zvenet’/prozvenet’ ‘ring, clank’ have intersection rates of 55, 51 and 50 respec-
tively. This is very close to the lowest intersection rate among the speech verbs
(60 for šeptat’/prošeptat’ ‘whisper’). The other two pairs of verbs have much
lower intersection rates: 23 for the pair pet’/propet’ ‘sing’ and 15 for the pair
gudet’/progudet’ ‘buzz’. These two groups differ semantically. While the first
group designates non-agentive sounds involuntary produced by inanimate
objects, the last two verbs usually refer to the sounds voluntary produced by
people (pet’/propet’ ‘sing’) or mechanisms operated by people (gudet’/progudet’
‘buzz’). Therefore these two groups represent agentive and non-agentive
sounds. We will see how this difference in the meaning of the sound verbs af-
fects the constructional profiles of the pairs of verbs and how that in its turn has
an effect on the intersection rate of the pairs of verbs.
Intersection rates for seventeen pairs of verbs | 147
Both verbs gremet’ and progremet’ ‘resound’ can refer to either loud noise or
achieving fame. The former uses are marked in Appendix with the gloss ‘re-
sound’ and the latter uses are marked with the gloss ‘achieve fame’. However,
even though both meanings are available to both the imperfective and the per-
fective verbs, the distribution among the two submeanings is different. The
metaphorical submeaning ‘achieve fame’ is more important for the perfective
verb than for the imperfective. The imperfective verb gremet’ ‘resoundIPFV’ ap-
pears in two different constructions that express the metaphorical meaning: the
intransitive construction NPnom V (see (93)) and the locative construction
NPnom V {po + NPdat / na + NPacc} denoting the area where the subject is fa-
mous; see (94).
||
35 Rozočke and kozočke sound similar and rhyme in Russian.
148 | Constructional profiling and aspectual pairs
However these two constructions appear in the sample four and three times
respectively; therefore neither of them is included in the constructional profile
of the imperfective verb. Yet, both constructions exceeded the threshold of five
examples for the perfective verb progremet’ ‘resoundPFV’ and are found in eleven
and thirteen examples respectively. All three constructions that appear in the
intersection of the two verbs are constructions where the verb denotes the pro-
duction of a loud noise. The perfective verb progremet’ ‘resoundPFV’ is found in
three such constructions: the intransitive construction NPnom V, the locative
construction [NPnom] V {ADV / PRwhere + NPx} and the direct speech construc-
tion NPnom V «CL». All three constructions are also found among the examples
of the imperfective verb gremet’ ‘resoundIPFV’ and therefore are found in the
intersection of the two verbs. In addition to those three constructions, the im-
perfective verb also is attested in the instrumental construction NPnom V NPins
(see (95)), which is found in only two examples of the perfective verb, so this
construction only appears in the constructional profile of the imperfective verb
and is absent from the constructional profile of the perfective verb.
Unlike the pairs where imperfective and perfective verbs stress different foci in
their constructional profiles, the pair zvenet’/prozvenet’ ‘ring, clank’ have the
same focus: both verbs most frequently appear in an intransitive construction
referring to inanimate objects NPnom V. In addition to this shared construction
there are two constructions in the constructional profile of the imperfective verb
that are not found in the profile of the perfective verb and two constructions in
the constructional profile of the perfective verb that are not found in the con-
structional profile of the imperfective verb. The imperfective verb zvenet’ ‘ring,
clankIPFV’ appears in the instrumental construction NPnom V NPins and the in-
transitive construction describing human speech NPnom V. The perfective verb
is found in the direct speech construction NPnom V “CL” which as we have seen
earlier is more frequent among the verbs with the prefix pro-. It is also found in
an idiomatic construction ANUM NPnom V, which describes a bell (zvonok or
Intersection rates for seventeen pairs of verbs | 149
zvonoček) that metaphorically rings above one’s head in order to give first, se-
cond or last warning; see (96).
(96) Ešče v načale dekabrja 1991 goda prozvenel vtoroj zvonoček nad golovoj
Lidy.
‘At the beginning of December 1991 Lida received a second warning.
(lit. rang the second bell above Lida’s head)’.
Summing up, the verbs of non-agentive sound have less overlap with the mean-
ing of the prefix pro- compared with the speech verbs. These verbs are likely to
have different foci for an imperfective and a perfective verb in a pair. The per-
fective verb behaves more like a speech verb: for example perfective verbs are
more likely to appear in the direct speech construction NPnom V “CL”. This is
the result of the influence of the semantics of the prefix pro- present in the per-
fective. We will see even more of this effect when we turn to the agentive sound
verbs, which are even more likely to be interpreted as speech verbs.
The pair pet’/propet’ ‘sing’ presents a clear picture where the perfective verb
is strongly affected by the semantics of the prefix pro-. The constructional pro-
file of the imperfective verb pet’ ‘singIPFV’ is spread across five different construc-
tions. The most frequent here is the intransitive construction NPnom V (thrity-
five examples), followed by the transitive construction NPnom V NPacc (twenty-
five examples), and the rest of the examples are distributed among the intransi-
tive locative construction NPnom V {v + NPloc / na + NPloc} (eight examples),
the direct speech construction NPnom V “CL” (6 examples) and the thematic
construction NPnom V {o + NPloc / pro + NPacc} (5 examples). Out of this abun-
dance the perfective verb propet’ ‘singPFV’ chooses only two constructions: ex-
actly those which are strongly preferred by the prefix pro-. These are the direct
speech construction NPnom V “CL” (57 examples) and the transitive construc-
tion NPnom V NPacc (17 examples); see (97)-(98).
The strong effect of the prefix pro- seen in the perfective verb propet’ ‘singPFV’
can be explained by the fact that the imperfective verb pet’ does not have large
overlap in meaning with the prefix pro-. As we can see from its constructional
profile the imperfective verb pet’ ‘singIPFV’ primarily describes production of a
150 | Constructional profiling and aspectual pairs
The imperfective verb focuses on marking a constant state of buzzing using the
intransitive construction NPnom V; see (100). This construction is not present in
the constructional profile of the perfective verb, since the perfective is not able
to refer to a constant state.
In addition to the two constructions shared by the two verbs and the stative
intransitive construction described above, the imperfective verb gudet’ ‘buzzIPFV’
also participates in the intransitive construction that describes a person hum-
ming NPnom V and an idiomatic construction that refers to an ache in a body
part; see (101).
In addition to the two shared constructions, the perfective verb is also found in
one other construction: the impersonal construction Vimpers; see (102). This
construction is absent from the constructional profile of the imperfective verb.
Intersection rates for seventeen pairs of verbs | 151
Thus, the pair of verbs gudet’/progudet’ ‘buzz’ again gives us a good example of
an imperfective sound verb that is not primarily used to describe communica-
tion. However, combined with the prefix pro- it assumes the role of a speech
verb, which changes its compatibility and its constructional profile. Again the
imperfective verb gudet’ ‘buzzIPFV’ does not have large overlap with the meaning
of the prefix pro- and we see more effect of the prefix pro- in the distribution of
the prefixed perfective.
Three pairs among the seventeen pairs in this study can be characterized as
having an intersection rate of 0, i.e. having no intersection between the con-
structions in the constructional profiles of the imperfective and perfective verbs.
Two pairs bit’/probit’ ‘beat’ and tjanut’/protjanut’ ‘pull’ actually have a 0 inter-
action rate, while the pair vesti/provesti ‘lead’ has the intersection rate 33. How-
ever, as shown below, all 33 uses involve examples where both imperfective and
perfective verbs function as light verbs. Therefore lexical uses of the verbs vesti
and provesti ‘lead’ do not share any constructions. Thus all three pairs can ac-
tually be characterized as pairs with no interaction; see Table 67.
In section 5.3.1 we have seen that the imperfective verbs with the highest inter-
section rate are characterized by a low level of polysemy. By contrast the verbs
vesti ‘leadIPFV’, bit’ ‘beatIPFV’ and tjanut’ ‘pullIPFV’ have a high level of polysemy.
While the imperfective verbs with the highest intersection rate have between
one and three submeanings in Ožegov and Švedova’s dictionary (1992), the
imperfective verbs in this section have notably more submeanings. In the same
dictionary the verb vesti ‘leadIPFV’ has nine submeanings, the verb bit’ ‘beatIPFV’
has twelve and the verb tjanut’ ‘pullIPFV’ has seventeen submeanings. More poly-
152 | Constructional profiling and aspectual pairs
semous verbs are more likely to participate in more constructions. While the
imperfective verbs with the highest intersection rate have between four and
fourteen constructions for a verb, the verbs with the lowest intersection rates
are characterized by nineteen (vesti ‘leadIPFV’), thirty-eight (bit’ ‘beatIPFV’) and
forty-three (tjanut’ ‘pullIPFV’) constructions. Examples of imperfective verbs are
spread among different constructions; this decreases the probability of intersec-
tion with the constructions of correlate perfective verbs.
Let us turn to the three pairs explored in this section and start with the pair
vesti/provesti ‘lead’. These two verbs have an intersection rate of 33. However,
this rate is based on one construction only, the transitive construction NPnom V
NPacc where the verbs vesti and provesti ‘lead’ function as light verbs; see (103)-
(104).
The term “light verb,” usually attributed to Jespersen (1965, Volume VI: 117),
was introduced in reference to English constructions such as have a rest, take a
drive, give a sigh. Even though they follow standard grammatical structures for
verbs like have, take or give, from the semantic point of view, these construc-
tions cannot be seen as full predicates. A light verb only provides verbal support
for a noun, which carries the semantic weight. Since Jespersen’s establishment
of the term, light verbs have been studied in various languages, including Japa-
nese, Hindi, Urdu and Romance languages (see Butt 2010 and references where-
in). In Russian, studies have identified the following light verbs: brat’/vzjat’
‘take’, vesti/provesti ‘lead’, davat’/dat’ ‘give’, delat’/sdelat’ ‘do’, imet’ ‘have’,
nesti/ponesti ‘carry’, položit’/klast’ ‘put (down)’, prinimat’/prinjat’ ‘take/get’,
stavit’/postavit’ ‘put (up)’ (Mudraya et al. 2008, Babych, Hartley and Sharoff
2009). We can see that light verb constructions are marked for aspect, because
both imperfective and perfective counterparts of most light verbs (except for the
verb imet’ ‘have’, which does not have a perfective correlate) appear in light
verb constructions. The distribution of the imperfective vs. perfective light verb
is usually affected by the direct object in the construction: most nouns have a
strong preference for either an imperfective or a perfective light verb; see ves-
Intersection rates for seventeen pairs of verbs | 153
The perfective verb also appears in a directional construction, but for the perfec-
tive verb this construction is transitive - NPnom V NPacc {ADV /PRto + NPx} (see
(107)). As we have seen earlier, perfective verbs with the prefix pro- have an
overall preference for the transitive construction, so this difference is in agree-
ment with the semantics that is introduced by the prefix pro-. The transitive
directional construction is also found among examples of the imperfective verb
(see (108)), yet there it is only attested in two examples, so this construction
does not appear in the constructional profile of the verb vesti ‘leadIPFV’.
In addition the perfective verb provesti ‘leadPFV’ appears in two temporal con-
structions that denote spending time: the adverbial temporal construction
NPnom V NPacc ADV (see (110)) and the locative temporal construction NPnom
V NPacc {ADV /PRwhere + NPx} (see (111)).
The idea of spending a bounded amount of time conforms with the semantics of
the prefix pro-. The usual semantic schema of going ‘through a quantum’ here is
applied to time. As mentioned above, the prefix pro- is often used to measure
incremental direct objects and specifically periods of time.
Thus, we have seen that the light verbs vesti and provesti ‘lead’ are similar.
As for the lexical verbs, the perfective verb provesti ‘leadPFV’ shows typical fea-
tures of a perfective verb with the prefix pro- both in spatial and in temporal
uses. However due to the high level of polysemy of the imperfective verb, nei-
ther of the constructions in its constructional profile intersect with the construc-
tions in the constructional profile of the perfective verb. Hence, we can con-
clude that the lexical verbs vesti and provesti ‘lead’ have a zero intersection rate.
The constructional profiles of the verbs tjanut’ and protjanut’ ‘pull’ show us
two different sets of constructions. It is important to note that the verb tjanut’
‘pullIPFV’ has the most submeanings according to Ožegov and Švedova’s diction-
ary and its attested in the most constructions among all verbs in this study. As a
result its uses are spread among different constructions and the most frequent
construction for this verb is attested in eleven examples (cf. eighty-nine for the
verb kontrolirovat’ ‘controlIPFV’). The frequent constructions for the verb tjanut’
‘pullIPFV’ focus on the submeaning ‘have an inclination, long for’; see the imper-
sonal construction NPacc Vimpers k + NPdat that marks inclination for an activi-
ty as in (112) and the directive impersonal construction NPacc Vimpers {ADV
/PRto NPx} that denotes inclination for moving toward a landmark as in (113).
The other two constructions refer to people following each other as in NPnom
Vsja {ADV /PRto NPx} (see (114)) and to time moving slowly as in NPnom Vsja
[ADV] (see (115)). Both constructions mark the verb with the detransitivizing
suffix -sja.
Note that all the constructions of the perfective verb agree perfectly with the
semantics of the prefix pro-. When we pass an object from one subject to anoth-
er the object goes through the space between the two subjects. When we string a
rope, it covers the space between the two landmarks. And when we speak, as
mentioned several times in previous sections, the sound crosses the space be-
tween a speaker and a hearer. Thus, the uses of the perfective verb protjanut’
‘pullPFV’ and its constructional profile are modified by the prefix pro-. The uses
of the imperfective verb tjanut’ ‘pullIPFV’ are spread among different construc-
tions and do not share the dominant idea of going ‘through a quantum’ with the
156 | Constructional profiling and aspectual pairs
perfective verb protjanut’ ‘pullPFV’. As a result the intersection rate for these two
verbs is 0.
Let us now discuss the pair bit’/probit’ ‘beat’. Here again we see how the
imperfective and the perfective verbs focus on different portions of the meaning
network. All constructions in the constructional profile of bit’ ‘beatIPFV’ denote
physical aggression. By contrast, the verb probit’ ‘beatPFV’ focuses on breaking
through a barrier (like a wall or mountain) or functions as a sound verb refer-
ring to striking in order to give a signal. Note that this preference is in accord-
ance with the semantics of the prefix pro-: the prefix pro- marks going ‘through
a quantum’, where a quantum can be an actual physical barrier or a space
through which a sound travels.
Transitive constructions NPnom V NPacc appear in the constructional pro-
files of both verbs. However, it can be argued that we are dealing with several
transitive constructions that differ in their semantics. The first transitive con-
struction appears in the constructional profile of the imperfective verb bit’
‘beatIPFV’. It denotes physical beating (119a), which cannot be expressed by the
perfective verb probit’ ‘beatPFV’ (119b).
The fourth construction also refers to breaking through, however now the direct
object is the resulting path, not a barrier as in the previous two constructions.
Again we see that the imperfective verb bit’ ‘beatIPFV’ cannot be used in this con-
struction; see (122).
Intersection rates for seventeen pairs of verbs | 157
The only perfective transitive construction that can be easily transformed into
an imperfective is a construction that describes striking in order to give a signal
as in (123).
This construction is found in twelve examples of the perfective verb and only
two examples of the imperfective verb. As a result this construction does not
appear in the constructional profile of the imperfective verb and is not counted
in the intersection rate. This raises an interesting question. The striking con-
struction is the only construction where the verbs bit’ and probit’ ‘beat’ are in-
terchangeable. In all examples of this construction both the imperfective and
the perfective verbs can be used if we adjust the grammatical background ap-
propriately. Yet, among the 100 randomly selected examples of the verb bit’
‘beatIPFV’ there are found only two examples describing striking clocks. Using
this number as an approximation we can assume that about 2% of all uses of the
verb bit’ ‘beatIPFV’ refer to clocks or other objects striking. Can we actually say
that the verbs bit’ and probit’ ‘beat’ serve as aspectual pairs if there is one con-
struction where they are interchangeable, but occurrences of this construction
are extremely rare among the uses of the imperfective verb? This leads us direct-
ly to the problem of prefix variation as related to aspectual pairs, which is ex-
plored in the next section.
The term “prefix variation” in this study refers to a phenomenon where a Rus-
sian imperfective verb has several prefixed perfectives with different prefixes
that all serve as perfective correlates of the imperfective verb. Until recently
prefix variation was seen as rare phenomenon, however lately Janda and
Lyashevskaya (2011b, see also Janda et al. 2013) have shown that prefix varia-
tion is a much more frequent phenomenon than previously expected. Janda and
158 | Constructional profiling and aspectual pairs
Lyashevskaya show that 27% of all simplex imperfective verbs use more than
one prefix to form perfective partner verbs. Among the seventeen simplex im-
perfectives in this study, four verbs have more than one perfective partner in the
Exploring Emptiness database. These are the verbs pet’ ‘singIPFV’, vesti ‘leadIPFV’,
tjanut’ ‘pullIPFV’ and bit’ ‘beatIPFV’. Each of them has three perfective partners.
The verb pet’ ‘singIPFV’ has perfective partners vospet’, propet’ and spet’. The
verb vesti ‘leadIPFV’ is paired with the verbs povesti, privesti, and provesti. The
verb tjanut’ can be perfectivized as vytjanut’, potjanut’ and protjanut’. The verb
bit’ ‘beatIPFV’ has the verbs pobit’, probit’ and razbit’ as its perfective partners.
Remarkably, imperfective verbs with multiple perfective partners belong to
pairs that received a low intersection rate in this study. These are the three pairs
with lowest intersection rate, plus the pair pet’/propet’ ‘sing’ which has an in-
tersection rate of 23, also a low intersection rate. This correlation is not coinci-
dental. If an imperfective simplex verb has a widely dispersed network of sub-
meanings, this motivates prefix variation since each prefix can focus on a single
subset of the meanings of the imperfective.
As can be easily seen from the information in the EE database, different per-
fectives are associated with different submeanings of the imperfective verb. The
perfective vospet’ can only be used as a perfective partner of the verb pet’
‘singIPFV’ in the submeaning ‘glorify in songs or poetry’ (see (124)). This sub-
meaning shows the influence of the meaning of the prefix voz- ‘up’.
The perfective spet’ ‘singPFV’ can be used to denote producing musical or har-
monious sounds referring to either humans or birds. This perfective realizes the
meaning ‘once’ of the prefix s- which often happens with verbs of speech, e.g.
sostrit’ ‘make a witty remark’, skalamburit’ ‘make a pun’, sformulirovat’ ‘formu-
late’ (see also Janda and Lyashevskaya 2013 on the semantic tendencies of per-
fectives with the prefix s-). The perfective verb propet’ can be used in both types
of contexts, but is more frequent in contexts denoting production of sounds,
since the prefix pro- agrees well with production of a sound.
A similar distribution can be found for the perfectives of the verb vesti ‘lead-
IPFV ’. The perfective povesti serves as a perfective counterpart of the verb vesti
(125) Ljubuju drevesinu vedet ot syrosti, vysoxnuv že, ona staraetsja vernut’sja
v prežnie gabarity.36
‘Any wood deforms from dampness, but once dry it tends to return to
the previous state.’
The perfective verb privesti is interchangeable with the verb vesti ‘leadIPFV’ when
it denotes either actual leading towards a landmark (127) or metaphoric produc-
tion of a result (128). These contexts are in accordance with the submeaning
‘arrive’ of the prefix pri- (see more on the semantics of the prefix pri- in chapter
4).
||
36 Source: http://artzeichn.ru/praktika/esse/tvorchestvo-copy_9.html
37 Source: http://asaratov.livejournal.com/2336743.html
160 | Constructional profiling and aspectual pairs
prefix vy- ‘out’. The perfective verb potjanut’ and the imperfective verb tjanut’
‘pullIPFV’ are interchangeable when used for pulling (potjanut’ kogo-libo za rukav
‘pull smb by the sleeve’), attracting (ego potjanulo domoj ‘he longs to go home’),
weighing (slitok potjanul na pjat’ kilogrammov ‘the ingot weighed as much as
five kilograms’) or making the grade as (na manekenščicu ona ne tjanet ‘she
won't make the grade as a model’). All these contexts fit into the network of
submeanings of the prefix po-. The contexts of pulling can be interpreted as
manifestation of the delimitative submeaning ‘some’ of the prefix po-. Attrac-
tion can be viewed as evidence of the ingressive submeaning ‘begin’. Both
weighing and making the grade as display the ‘result’ submeaning of the prefix
po-. The perfective verb protjanut’ functions as a perfective partner of the verb
tjanut’ ‘pullIPFV’ when it refers to making via drawing (protjanut’ provoloku ‘make
a wire’), pulling (protjanut’ ruku ‘reach out one's hand (for smth)’), stringing a
rope (protjanut’ verevku ‘stretch the rope’), talking slowly, delaying or procras-
tinating (protjanut’ s otvetom ‘procrastinate with one's answer’) and surviving
using scarce resources (protjanut’ nedelju na odnom kompote ‘survive one week
eating only compote’). We have seen that many of these meanings appear
among the constructions available for the perfective verb provesti and agree
well with the semantics of the prefix pro-, however these submeanings are in-
frequent among the uses of the imperfective verb tjanut’ ‘pullIPFV’, so neither of
them appears in the constructional profile of the imperfective verb tjanut’ ‘pul-
lIPFV’, and as a result the two verbs have an intersection rate of 0.
The perfective verb pobit’ is the most neutral among all perfective partners
possible for the verb bit’ ‘beatIPFV’. It is mostly used in the contexts of physical
beating. These contexts are in accordance with the resultative meaning of the
prefix po-. The verb razbit’ functions as an aspectual partner of the verb bit’
‘beatIPFV’ in the contexts describing the breaking of a fragile object into small
pieces (razbit’ okno, čašku ‘break the window, cup’), which agrees well with the
semantic schema of the prefix raz- ‘apart’. As mentioned in the previous section,
the perfective probit’ serves as a partner for the imperfective bit’ ‘beatIPFV’ in
contexts denoting designating something by sonorous strokes, mainly used to
refer to clocks, bells or cannons that sound to indicate time; see (129). As dis-
cussed earlier this submeaning is infrequent among the uses of the imperfective
bit’ ‘beatIPFV’, which results in an intersection rate of 0 for the pair bit’/probit’
‘beat’.
(129) Probila puška. No èto vovse ne značit, čto polden’. [Mixail Šiškin. Vener-
in volos (2004) // «Znamja», 2005]
‘The cannon fired. But it does not mean that it is noon.’
Intersection rates for seventeen pairs of verbs | 161
We see an obvious pattern in the case when various perfectives serve as perfec-
tive partners of the same imperfective verb. This usually happens when the
imperfective has a widely dispersed semantic network with numerous mean-
ings. Some of these meanings are more compatible with the meaning of one
prefix, while other meanings are more compatible with another. As a result we
see that different submeanings choose different perfective partners. Janda et al.
(2013: 162) summarize similar findings based on a larger set:
Prefix variation is clearly governed by the meanings of the prefixes involved. There are
some simplex verbs that can show overlap with more than one prefix, and these verbs
tend to cluster in groups according to their meanings.
This brings us back to the original question. What does it mean for two verbs to
be aspectual partners of each other? Does the existence of the pair pet’-spet’
‘sing’ present a problem for the pair pet’-propet’ ‘sing’? It is important to note
here that all criteria for establishing aspectual pairs are based on context. The
definition of the aspectual pairs based on different contexts or different sub-
meanings of the imperfective is very close to the definition via constructions.
Let us assume that a child acquiring Russian hears an imperfective and a per-
fective in similar contexts. How does the child realize that the given contexts are
similar? He or she can recognize a pattern and relate that pattern to an imper-
fective and a perfective. If we accept that verbs like bit’ and probit’ ‘beat’ may
function as aspectual pairs of each other, even though they are interchangeable
only in one particular context, which is infrequent for the imperfective verb,
then we are assuming that aspectual pairs are defined on a basis of a single
construction.
Summing up the section on imperfectives and their perfective correlates
with the prefix pro-, we can see that perfectives with the prefix pro- share sever-
al tendencies. The perfectives are more narrow in their use than the correlated
imperfectives: overall the constructional profiles of the perfectives contain few-
er constructions, while each of the attested constructions has more examples
(with several exceptions discussed in description of Table 59 in section 5.2).
Perfectives with the prefix pro- have a preference for several constructions. The
prefix pro- means ‘through a quantum’ and adds to perfectives a preference for
contexts where the situation includes an incremental object and “physical ex-
tent of the Increment directly influences the duration of an event” (Paducheva
and Pentus 2007: 203); see examples (130) and (131) where all the Potter books
and letter serve as incremental objects of the verb procitat’ ‘readPFV’.
162 | Constructional profiling and aspectual pairs
(130) Za èti šest’ nedel’ pročitala vsego Pottera... [Naši deti: Podrostki (2004)]
‘During these six weeks she read all the Potter books.’
Perfectives also have a preference for transitive constructions. This is partly the
result of a preference for incremental contexts and partly an overall perfective
preference for explicit objects. Perfectives with the prefix pro- also have prefer-
ence for the direct speech construction. This preference is in agreement with the
general attraction of the prefix pro- towards speech verbs, which express the
traveling of sound waves through space between speaker and listener. The di-
rect speech construction is attested even with perfective verbs where imperfec-
tive correlates are not compatible with direct speech, such as kontrolirovat’
‘controlIPFV’; see (134).
A single clear picture emerges from the tendencies shown by perfective verbs
with the prefix pro-. These tendencies result from the meaning of the prefix pro-.
||
38 Examples (132) and (133) repeat examples (53) and (55) cited earlier in this chapter.
Summary | 163
The imperfective verbs resemble their perfective correlates with the prefix pro-
as long as the semantics of the imperfective overlaps with the meaning of the
prefix pro-.
5.4. Summary
semantics of the prefix pro-, the more likely it is that the imperfective and the
perfective will use a different sets of constructions.
Overall it can be shown that the production of prefixed perfective correlates
from simplex imperfectives does not differ notably from the production of any
other verbs derived from that simplex imperfective. We can say that when a
native speaker of Russian is challenged with the task of modifying a verb in a
given sentence to receive a specified additional grammatical meaning, this can
be seen as if all native speakers have access to a function v2 = F(v1, C, G), where
v1 refers to the verb in the original sentence, C to the context where the verb v1 is
used and G to the new grammatical meaning that the speaker wants to express.
The formula v2 = F(v1, C, G) means that a native speaker of Russian, when given
a verb, a context where the verb is used, and a new grammatical meaning that
he or she wants to add to the sentence, can produce another verb v2 that ex-
presses this intended grammatical meaning and lexical meaning of the original
verb v1 in the context C. Applying this to the prefixal aspectual pairs we can use
a perfective verb as v1, a context where the perfective verb is used as C, and any
of the grammatical contexts proposed for establishing a paired imperfective, for
example praesens historicum, as new grammatical meaning G. As a result we
will receive an imperfective v2 that functions as an aspectual partner of the per-
fective v1 in the context C. For example, for the perfective v1 pročitat’ ‘readPFV’ in
the context C set by the sentence (135) and praesens historicum as grammatical
meaning G, the result of the function F(v1, C, G) is v2, čitat’ ‘readIPFV’; see (136).
(135) Ja pročital knigu Karpova “100 lučšix partij” i posmotrel po baze dannyx
kakie debjuty Anatolij Evgen’evič igraet v poslednee vremja. [Darmen
Sadvakasov. Izučil sto, sygral vosem’ (2004) // “64 – Šaxmatnoe oboz-
renie”, 2004.11.15]
‘I have read the Karpov's book “100 best games” and looked in the da-
tabase at what openings Anatoly Evgenievich has played recently.’
(137) Abduraxman sidel vozle okna i čital knigu. [Čukotskie očki // “Tramvaj”,
1990]
‘Abdurahman was sitting near the window and reading a book.’
(139) Japonec v otvet kak-to daže isteričeski kričit: "Ja ― japonec!.." [Zapis’
LiveJournal (2004)]
‘The Japanese in response to something almost hysterically cries: "I am
a Japanese! .."’
(141) V kvartire, kuda zašel Pal’to, zvuki byli obyčnymi dlja sobirajuščegosja
zakusit’ čeloveka: stavilsja čajnik na plitu, xlopala dverca xolodil’nika,
zvjakala posuda. [Vladimir Čerkasov. Černyj jaščik (2000)]
‘In the apartment, where the Pal’to went, the sounds were normal for a
person who is about to eat: the kettle was put on the stove, the refrig-
erator door was slammed, dishes were clinking.’
Thus, it seems that there is nothing particularly special about native speakers of
Russian being able to produce an aspectual partner for a given verb. There are
many kinds of perfectives that can appear in the same contexts as the imperfec-
tive. With the same ease native speakers of Russian can produce ingressives,
delimitatives, resultatives, semelfactives and make other grammatical modifica-
Summary | 167
tions to a given verb in a given context. It is not clear why linguistic research on
Russian aspect is so focused on aspectual pairs. Aspectual pairs are a product
of structuralist thinking and presuppose key ideas of structuralism such as
binary oppositions, neutralization and substitution.
The reasoning above draws together the perfective aspectual correlates and
other kinds of perfectives (Specialized Perfectives, Complex Act Perfectives,
Single Act Perfectives in terms of Janda 2007). All kinds of perfectives seem to
have a similar origin and are derived via similar mechanisms. This notion is in
agreement with the Overlap hypothesis. One of the consequences of this hy-
pothesis is that perfective aspectual correlates are not so different from the oth-
er kinds of perfectives and the mechanism described above shows that cogni-
tively their production could be described in a way similar to that for the
production of other kinds of perfectives.
6. Collostructional profiling
Similar to constructional profiling discussed in the previous chapter, collostruc-
tional profiling uses Construction Grammar as its theoretical background and
investigates the relationship between the construction and its elements. How-
ever, while constructional profiling focuses on the list of constructions available
for a lexeme, collostructional profiling is a method that allows us to reveal the
semantic restrictions on a constructional slot. It compares the distribution of
fillers of a slot in the construction with the overall distribution of the same ele-
ments in the corpus and shows whether semantic preferences affect the distri-
bution significantly. This chapter is structured as follows. Section 6.1 discusses
statistical methods offered for exploration of a slot of a construction, mainly
collostructional analysis. The mechanism of collostructional profiling is out-
lined in the beginning of section 6.2. Then the remaining part of section 6.2
discusses in more detail preferences placed on verbal slots by prepositional
phrases (PPs) with the word ‘phone’ in both Russian and English. The next
section (6.3) describes what features of the situation of speaking into the phone
attract verbs of shouting and whispering. The last section (6.4) describes how
collostructional profiling can be used to probe the interaction of two construc-
tions. This case study shows what happens in the interaction of the Russian
possessive construction and the construction of disappearance. This section
also illustrates how collostructional profiling can be combined with collostruc-
tional analysis39.
||
39 A shorter version of this chapter was published as Kuznetsova 2010; however, this chapter
presents a more detailed description of the research and more thorough discussion of the
results than the previously published work.
Grammatical tendencies and collostructional profiles | 169
(141) «What?» - she said into the phone. [John G Hemry. Betty Knox and
Dictionary Jones in The Mystery of the Missing Teenage Anachronisms
(2011) Analog Science Fiction & Fact. Vol. 131, Iss. 3]
(142) Simon was frowning as he spoke into the phone. [Shay, Kathryn. Be my
babies (2008) Toronto, Ont. ; New York : Harlequin]
(143) Pretend you're sick - here, cough into the phone. [Sutton, Barbara. The
Brotherhood of Healing (2002) Antioch Review. Summer2002, Vol. 60
Issue 3]
(144) He sat in a chair and leafed through one of Bubbles' Bombay fan mags
while she cooed into the phone. [Sacred Cow (1993 (Jan)) Omni. Vol. 15,
Iss. 4]
This method enables us to say what classes of words are compatible and not
compatible with a construction in question. However lists of compatible and not
compatible classes do not fully describe how semantic restrictions on a slot
work, because semantic restrictions are also governed by tendencies. Some of
the compatible words are more preferred by a slot, while other words that are
also possible in a slot are dispreffered by a construction. In section 6.2 we will
170 | Collostructional profiling
see that the into the phone PP has a preference for verbs of shouting and whis-
pering.
Semantic restrictions on a construction slot can be compared with shaped
filters as in Figure 23. The items that suit the form of the filters best are easily let
through, while items of other shapes have difficulties coming through.
The ‘mutual information’ (MI) measure is based on effect size, whereas the z-
score, t-score, and simple-ll are based on significance. The MI is known to be
low-frequency biased; it gives high effect size estimates for a small number of
observations. The latter measures are known to be high-frequency biased: for
many observations they report high significance and a small relative distance
between the observed and expected frequencies (i.e. small effect size estimates)
(cf. Evert 2008).
Some of the limitations associated with these simple measures can be
solved via contingency tables. For example, let us investigate how often the
Russian verb pet’ ‘sing’ co-occurs with the PP po telefonu ‘into the phone’ using
the Russian National Corpus (RNC). In order to use a contingency table we need
to collect the number of examples attested for the co-occurrence of these two
items. This information appears in the top row of the second column of Table
68. We see that pet’ po telefonu ‘sing on the phone’ occurs only twice in the
whole RNC. Then, we need to know how often each of the items occurs without
the other. We see that other verbs occur with the PP po telefonu ‘on the phone’
2,342 times, whereas the verb pet’ ‘sing’ without the PP po telefonu ‘on the
phone’ is attested in 486,464 examples. Finally, for the last slot in the contin-
gency table we need the number of the remaining items. Here, we have verbs
other than pet’ ‘sing’ co-occurring with PPs other than po telefonu ‘on the
phone’ in 4,511,806 examples.
Table 68: Contingency table of observed frequencies for co-occurrence of the verb pet’ ‘sing’
and PP po telefonu ‘on the phone’
Several measures can be used for the statistical analysis of such contingency
tables. Linguists most frequently use the chi-square test (with and without
Yates’ continuity correction), log-likelihood measure, and Fisher’s exact test;
see Evert (2004: 54–58) who discusses the mathematical properties of these
measures.
S. Gries and A. Stefanowitsch have developed a statistical approach they
call “collostructional analysis”, which measures the attraction and repulsion of
a lexeme for a slot of a construction using one of these measures – Fisher’s ex-
act test (Stefanowitsch and Gries 2003, 2005, Gries and Stefanowitsch 2004, and
others). For example, Stefanowitsch and Gries (2005) discuss which lexemes are
attracted and repulsed in the causative into-construction. Using the frequencies
Grammatical tendencies and collostructional profiles | 173
of two lexemes filling different slots in the construction (for example fool and
thinking), they can predict what frequency their pairing would have if these
events were independent. Comparing that prediction with the actual frequency
of the pair, the authors make a conclusion about the attraction or repulsion of
the two lexemes in the construction. Two lexemes are attracted if the actual
frequency is higher than the prediction. Two lexemes are repelled if the actual
frequency is lower than the prediction. For example, Stefanowitsch and Gries
show that fool into thinking occurs much more frequently than fool into V-ing
and V into thinking would predict.
The semantics of some frames coincides with the semantics of the construc-
tion and such elements of the frames are attracted to the construction, while
some pairs of verbs do not constitute a suitable frame and as a result are re-
pulsed from a construction. Tricking somebody into believing something is well
entrenched in the mind of speakers of English and therefore the examples of
this frame such as fool into thinking or mislead into believing appear at the top of
Stefanowitsch and Gries’ list of attracted lexemes. On the other hand physical
aggression is an ineffective way to change someone’s mind, and as a result we
see that items reflecting this frame such as force into thinking or bully into believ-
ing are repulsed from the construction. Thus collostructional analysis uncovers
the semantic structure of a construction. However collostructional analysis has
several disadvantages, most of which can be avoided if we use collostructional
profiling.
Bybee (2010: 97-101) offers several critiques of collostructional analysis and
Gries (2014: 24-35) answers to some of them. In the following, I highlight points
from this debate that are important for this chapter. First, Bybee draws attention
to the fact that a word that is frequent overall can also be a central member of a
category for a given construction. In this situation, the collostructional analysis
will not include such words in the list of the items attracted to the construction-
al slot. Highly frequent words such as do, talk, or walk do not appear on the list
of attracted lexemes in collostructional analysis, since they are usually not at-
tracted to a construction with a specific meaning, such as for example causa-
tion. For instance, if we consider again the PP into the phone, the verb talk is not
in the list of the verbs attracted to the PP on the phone. Moreover, it would be in
the list of items repulsed from the PP into the phone. This is a minus because
even though these frequent verbs are not attracted to a slot, among all examples
of a construction they do appear frequently due to their overall frequency. As
we know from experimental studies conducted by Goldberg (2006), the items
that appear in a slot frequently contribute to our understanding of a construc-
tion. The most frequent items appearing in a slot give us information about the
most neutral possible filler for the slot. These filler words usually indicate the
174 | Collostructional profiling
semantic class allowed in a slot, so by removing focus from these words we lose
important information about semantic restrictions on the slot.
Gries (2014: 25) discusses this issue using an English construction with the
predicative as as an example:
(146) I definitely don't see myself as a hero. [What are Rick Perry's political
and cash connections? 2011 CNN_Cooper]
Gries, Hampe, and Schönefeld (2005) analyzed this construction and show that
the verb regard has the highest collocation strength in this construction. Gries
points out that if Bybee’s logic were sound, the predicative as construction
would be better characterized by the verbs see and describe, which are the most
frequent verbs in this construction, than by the verb regard. Gries (2014: 25)
argues that this result would be “unintuitive”. However, it might be the case
that exactly the verbs see and describe shed the most light on the meaning and
distribution of the predicative construction. Kuznetsova and Rakhilina (2010)
explored Russian depictive construction, which can serve as an analogue for
English predicative construction. The Russian verbs kazat’sja ‘seem’ and sčitat’
‘consider’, which are among the most frequent fillers of the verbal slot of the
depictive construction, are frequently excluded from analysis of the depictive
construction. Kuznetsova and Rakhilina demonstrate that examples that use
these verbs are among the most prototypical uses of the depictive construction,
and they argue that these verbs best characterize the depictive construction.
Their findings suggest that the verbs see and describe might also play an im-
portant role in the use of the English predicative construction. Collostructional
profiling represents an improvement over collostructional analysis because it
does not discriminate against the most frequent verbs: these verbs are usually
part of the collostructional profile of a slot even if they are dispreferred by the
semantic restriction of the slot.
Second, Bybee notes that one of the factors used by Stefanowitsch and Gries
in the Fisher test – the number of all constructions in the corpus – is problemat-
ic, since “[t]here is no known way to count the number of constructions in a
corpus because a given clause may instantiate multiple constructions” (Bybee
2010: 98). Schmid and Küchenhoff (2013: 542-544) also mention that the idea of
the frequency score of all constructions and all lexemes other than investigated
is controversial, because it “depends on a subjective decision made on the basis
of linguistic theorizing”. Gries (2014: 26) argues that quantitative studies of
collocations traditionally use “the number of lexical items in the corpora” or
“the number of all verbs” as the estimate for the fourth slot in a contingency
table. Gries points out that this approach has been shown to yield linguistically
Grammatical tendencies and collostructional profiles | 175
Table 69: Collostructional profiles of the first and the second variable of the construction X of Y
(first 25 elements)
X occurrences Y occurrences
kind 233 people 64
lot 211 life 62
number 179 water 50
sense 105 women 40
lack 67 time 38
level 62 children 30
178 | Collostructional profiling
X occurrences Y occurrences
amount 53 power 30
piece 52 music 29
couple 48 education 28
variety 48 students 27
type 42 things 27
percent 39 war 27
group 38 information 26
professor 36 others 23
use 35 research 23
kinds 34 interest 21
years 34 money 21
set 33 government 20
levels 30 work 20
department 29 violence 19
range 29 $ 18
series 29 health 18
thousands 28 land 18
source 27 light 18
line 25 food 17
Yet cup does not signal a type and usually is not meant to be a measure when
one talks about a cup of tea, and tea is certainly neither a person nor an abstract
noun. So we can see that an idiom, which according to collostructional analysis
is most strongly attracted to the of-construction, does not provide much infor-
mation about the slots of the construction.
To conclude this comparison, I would like to add that collostructional anal-
ysis and collostructional profiling give the best results when both are used to-
gether: the list of items and conclusions about overall constructional tendencies
can be deduced using collostructional profiling, while the contribution of each
item in the list can be studied using collostructional analysis. The combination
of the two methods will be discussed in section 6.5 using the construction of
disappearance and the possessive construction in Russian.
Another similar statistical method of linguistic analysis was proposed by
Stubbs (1995). Stubbs explores what he calls “semantic prosodies of words”,
which include lists of individual collocations of the word in a corpus (Stubbs
also uses the term “semantic profiles” for semantic prosodies, but since the
term “semantic profiles” is used for another notion in this book, I will use the
term “semantic prosodies” to avoid confusion). The list of collocations provides
interesting insights into the meaning of the word in interest. For example
Stubbs demonstrates that the word cause is associated with negatively evaluat-
ed words, and frequently collocates with such words as abandonment, accident,
Collostructional profiling of phone PPs in Russian and English | 179
Many statistical calculations compare (a) how often something is actually observed and
(b) how often it might be expected merely by chance. And, given how experiments are de-
signed, one is usually hoping that (a) is much bigger than (b). A comparison of observed
and expected frequencies of pairs of words usually starts from the assumption that it is
meaningful to compare (a) a real corpus and (b) a hypothetical corpus consisting of the
same words in random order. We then look for statistically significant deviations from this
hypothetical randomness. The conventional null hypothesis would be that there is no dif-
ference between the real and hypothetical corpora.
This assumption is problematic because language data never has random dis-
tribution, therefore true independence of variables can rarely be observed (see
also Kilgariff 2005, which is characteristically entitled “Language is never, ever,
ever, random”). However, if we are comparing a frequency list for a slot with the
distribution in a corpus, we do not need to assume random distribution in a
corpus: we are taking the distribution in the corpus as a point of departure and
examine whether that distribution is significantly changed in the pattern of
interest. Thus, collostructional profiling is similar to semantic prosodies pro-
posed by Stubbs, but combines it with the framework of Construction Grammar.
As a result we can, first, narrow the search for the relevant fillers based on the
structure of the construction, and second, explain why the resulting list of
words shows the semantic similarities noticed by Stubbs. In Construction
Grammar, these similarities are an effect of semantic restriction on the slot of
the construction.
3. Collect all examples of the construction from the corpus; in order to fulfill
the independence assumption we need to collect only one example per au-
thor.
4. Exclude words that appear less than five times in a studied slot. This deci-
sion is somewhat arbitrary. However it is necessary to draw the line be-
tween the more frequent and less frequent fillers of the slot, since less fre-
quent elements are worse predictors of the requirements of the slot. The
choice of a threshold of five items is motivated for two reasons. On the one
hand, use of the chi-square test is appropriate only if the expected frequen-
cies are more than five. There is certainly no guarantee that an item that has
an observed frequency of five will also have an expected frequency of five.
However for an item that has even lower observed frequency, an expected
frequency of five is even more unlikely. On the other hand, all case studies
in this chapter show that a threshold of five gives reasonable predictions:
we receive neither too many nor too few items to analyze.
5. Find how often the same items are used overall in the corpus.
6. Measure the difference between the frequencies of the fillers in the con-
struction and the distribution of the same words in the corpus, using Pear-
son’s chi-square test and effect size measure Cramer’s V. In calculating both
Person’s chi-square and Cramer’s V we test an array of sample frequencies
against a prespecified set which shows the frequencies predicted by the null
hypothesis. In our case the null hypothesis distribution is the distribution
similar to the distribution in the corpus, and the sample distribution is the
distribution in the construction of interest.
This case study explores English and Russian speech verbs with PPs, containing
‘phone’. It investigates two hypotheses: 1) A phone PP produces an independent
construction and 2) A phone PP can be freely added to any speech verb. Two
constructions in English and two constructions in Russian are used as the mate-
rial for the analysis. In both languages I explore the most generally used phone
PP and compare it with a PP meaning ‘into the phone’. This case study shows
that the English on the phone phrase can be added almost freely to any speech
and sound verb, while other phone PPs produce phone constructions with sig-
nificantly different semantic constraints.
In both languages I investigate the most frequently used PPs (on the phone
in English (147) and po telefonu ‘on phoneDAT’ in Russian (150)) and compare
them with the PPs with the preposition ‘into’. While the most frequent variants
Collostructional profiling of phone PPs in Russian and English | 181
are never used with the ‘receiver’ (neither *talk on the receiver, nor *govorit’ po
trubke ‘talk on the receiverDAT’ are found in the corpora), the ‘into’ PP allows the
receiver as a possible filler in both languages, so I use two variants of fillers
with the preposition ‘into’: ‘phone’ and ‘receiver’. This gives us into the phone
(148) and into the receiver (149) in English and v telefon ‘into the phoneACC’ (151)
and v trubku ‘into the receiverACC’ in Russian (152)).
(147) It feels wrong to sit in my pajamas talking on the phone with a U.S.
attorney in D.C., sounding tough about a criminal he's trying to put away.
[Huston, James W. Marine One (2009)]
(148) One guy called up and just screamed into the phone, no contaminated
blood! [One-Horned Unicorn Deer Found in Italy. The Bryant Park Pro-
ject 8:00-9:00 AM. (2008)]
(149) "I'm here," Harry said into the receiver, then listened without a word.
[Englert, Jonathan. A dog among diplomats :a Bull Moose Dog Run
mystery. New York, N.Y. : Dell Book (2008)]
We will see that the distribution of the two fillers is different in Russian and
English. While English more frequently uses the into the phone variant, Russian
shows preference for v trubku ‘into the receiver’. The fact that the receiver can be
used with the preposition ‘into’, but cannot be used in the neutral phone PP can
182 | Collostructional profiling
be explained. In section 6.4 I will discuss why talking into the phone or receiver
in both English and Russian has to be an observable event. In this event the
phone functions as a physical object and can be replaced with a receiver. How-
ever, talking on the phone in English or po telefonu ‘on phoneDAT’ in Russian
describes functional use of the phone, where the phone cannot be replaced with
a receiver.
The structure and typical examples of these PPs are given in Table 70. For
the purposes of this paper I consider a verb to be a speech verb if it describes a
sound that can come out of a person’s mouth. For example, I explore such verbs
as breathe or sigh that are usually not considered to be speech verbs. Verbs that
denote acts of communication such as say, speak or talk are referred to as neu-
tral speech verbs and are opposed to the verbs that introduce some additional
information about the character of communication such as English shout and
whisper or Russian zagovorit’ ‘start talking’.
PP examples
English 1) on + NP speak on the phone
2) into + NP speak into the phone
speak into the receiver
Russian 3) po + NPdat govorit’ po telefonu ‘speak on
phone-DAT’
4) v + NPacc govorit’ v telefon ‘speak into
phone-ACC’
govorit’ v trubku ‘speak into
receiver-ACC’
Table 71 shows how many occurrences of each phone PP were found in the cor-
pus. A speech verb that appears with a PP five or more than five times is includ-
ed in the list of top speech verbs for that PP. Table 71 also shows how many top
speech verbs are found for every phone PP. It can be seen that usually there are
eight or more top speech verbs for a PP, but the English PP into the receiver and
the Russian PP v telefon ‘into the phone’ have noticeably fewer favored speech
verbs – four and five respectively. To make the data for the English into NP PP
the Russian v NPacc PP more comparable with data for other phone PPs, I ex-
plored in more detailed the most frequent fillers of both PPs: into the phone for
English and v trubku ‘into the receiver’ for Russian.
Collostructional profiling of phone PPs in Russian and English | 183
This section applies collostructional profiling to the speech verbs with the on
the phone PP in English. All verbs that appear in the context of on the phone in
CoCA are collected (7230 examples). Table 72 shows the eight verbs that ap-
peared more than four times in this small subcorpus. The column labeled “cor-
pus” shows how many examples of this verb are found in the corpus. The col-
umn labeled “predicted” gives us the number of examples that would occur
before the PP if that distribution were similar to the distribution in corpus. The
column labeled “observed” shows how many examples of that verb are found in
the context of PP on the phone. The numbers in the column “predicted” are
calculated using the following formula. The eight speech verbs that are used
most frequently with into the phone are taken (990 examples, see row “TOTAL”).
The number of occurrences for the same verbs in the corpus is calculated
(390,370 examples).
||
40 In Table 72 and similar tables below expected raw frequencies are rounded to integers.
However for some verbs (e.g the verb gab in this table) the expected frequency is 0, so the
frequency rounded to three digits is shown in brackets.
184 | Collostructional profiling
For each verb the percentage of its occurrences in the corpus is calculated. For
example, for the verb talk, which occurs in 256,892 examples in the corpus, this
percentage is 65%, since 256,892 is 65% of 390,370. Thus if the distribution with
on the phone were the same as in the corpus it would occur in 651 examples
(65% of 990 examples), however actually it occurs in 741 examples, as can be
seen from the column labeled “observed”. Both Pearson’s chi-square and
Cramer’s V effect size index are calculated using the predicted and observed
probabilities.
The semantic field of speech and sound in CoCA is dominated by two verbs:
talk and speak, as the left pie chart of Figure 24 shows. The same verbs domi-
nate with the phone PP on the phone, as can be seen from the right pie chart of
Figure 24. The two charts show that the distribution of the speech and sound
verbs with on the phone is relatively the same as the distribution of these verbs
in the corpus. While the chi-square test shows that the difference is statistically
significant (χ2 = 37.4633, df = 7, P = 3.832e-06), the effect size index Cramer’s V =
0.14 shows that the size of the effect is small. Thus the on the phone PP is closer
to the prediction of the second hypothesis: the English on the phone phrase can
be attached almost freely to any speech verb. The most frequent speech verbs
are frequent in this construction and vice versa. Therefore the on the phone PP
does not add a lot of specific information and does not pose additional semantic
requirements on the verbs used with it.
If we look at a similar table for the into the phone PP, we see that the distribu-
tion in the corpus is similar to the previous case dominated by neutral speech
and sound verbs: say, speak and talk. Thus the distribution in the CoCA predicts
Collostructional profiling of phone PPs in Russian and English | 185
that these verbs should dominate the distribution for the into the phone PP. The
column labeled “prediction” shows a prediction of 209 occurrences of say, thir-
teen occurrences of speak, thirty occurrences of talk and three or fewer occur-
rences of other verbs. However, the distribution of the top speech and sound
verbs with into the phone is noticeably different. Neutral speech verbs such as
say and talk appear less frequently than predicted, while verbs of shouting
(scream, shout, yell, bark) and verbs of whispering (whisper, sigh, breathe) ap-
pear more frequently than the corpus predicts.
Table 73: Top speech verbs with the PP into the phone
verb corpus into the phone: predicted into the phone: observed
say 1,845,675 209 96
speak 112,668 13 43
scream 21,312 2 29
whisper 18,640 2 22
shout 19,045 2 19
yell 13,531 2 18
talk 262,293 30 17
sigh 13,433 1 9
breathe 23,673 3 7
bark 6,675 1 5
TOTAL 2,336,945 265 265
The differences between the pie chart on the left of and on the right of Figure 25
show that the distribution of the speech and sound verbs with into the phone is
different from the distribution of the same verbs in the corpus. This difference is
statistically significant (χ2 = 138.9441, df = 9, P < 2.2e-16), and the size of the
effect is large (Cramer’s V = 0.51). Thus for this PP the first hypothesis is con-
firmed: it produces a special phone construction with specific semantic re-
quirements on the verb that can be used in it.
186 | Collostructional profiling
Fig. 25: Top speech verbs with the PP into the phone
As can be seen from Table 74, the into the receiver PP in English is rare in the
corpus. Its collostructional profile contains only four verbs. However, even
though these four verbs do not give us enough information for a statistical anal-
ysis, we see that the overall distribution of speech verbs with into the receiver is
similar to the distribution with into the phone. In addition to the neutral verbs,
say and speak, the collostructional profile contains the verbs whisper and shout.
These verbs represent the classes of shouting and whispering verbs that are
attracted to the into the phone PP. We will see that for the Russian PPs v telefon
‘into phoneACC’ and v trubku ‘into receiverACC’, the distribution is similar: the
collostructional profile of the verbal slot with the more rare filler looks like a
truncated version of the collostructional profile of the same slot with the more
frequent filler. However in Russian v trubku ‘into receiverACC’ is the frequent
filler and v telefon ‘into phoneACC’ is the rare filler.
Table 74: Top speech verbs with the PP into the receiver
verb corpus into the phone: predicted into the phone: observed
say 1,845,675 36 15
speak 112,668 2 9
whisper 18,640 0.364 8
shout 19,045 0.372 7
TOTAL 1,996,028 39 39
If we turn to the Russian analogue of the English on the phone PP, we see that
the top speech verbs with the po telefonu ‘on the phone’ PP mostly include the
Collostructional profiling of phone PPs in Russian and English | 187
neutral speech verbs; see Table 75. We have seen that in English the on the
phone PP can be added almost freely to any speech and sound verb. As opposed
to that here the observed numbers of occurrences are noticeably different from
those that are predicted.
Table 75: Top speech verbs with the PP po telefonu ‘on the phone’
Figure 26 shows that the distribution of the speech and sound verbs with po
telefonu ‘on the phone’ is different from the distribution of the same verbs in the
corpus. This difference is statistically significant (χ2= 361.7146, df, = 8, P < 2.2e-
16), and the size of the effect is large (Cramer’s V = 0.54). The main difference is
in the distribution of the most neutral verbs govorit’ ‘talk’ and skazat’ ‘say’.
While the verb skazat’ dominates in the corpus, the po telefonu PP is dominated
by govorit’ ‘talk’.
This difference is affected by the aspect of an event. The nature of the situa-
tion of speaking on the phone presupposes that the situation lasts over a period
of time. As a result the verbs that denote protracted, durative events such as
govorit’ ‘talk’ are preferred by this PP, while delimited and punctual events such
as skazat’ ‘say’ are dispreferred. Because of this preference all imperfective
speech verbs (govorit’ ‘talk’, razgovarivat’ ‘converse, sprašivat’ ‘ask, boltat’
‘chatter, rasskazyvat’ ‘tell, orat’ ‘yell’) and the durative perfective pogovorit’
‘talk for a while’ are used with the po telefonu ‘on the phone’ PP more frequently
than the corpus predicts. The punctual perfective verb in the list, skazat’ ‘say’,
is used less frequently than overall. Thus the po telefonu ‘on the phone’ PP pro-
duces a new independent phone construction that is sensitive to the aspect of
the event.
188 | Collostructional profiling
Fig. 26: Top speech verbs with the po telefonu ‘on the phone’ PP
6.2.4. V telefon ‘into the phone’ and v trubku ‘into the receiver’
There are only five speech verbs with five or more examples of v telefon ‘into the
phone’ PP in the RNC. However even this short list shows preferences similar to
those we observed for the English analogue into the phone. While neutral
speech verbs such as govorit’ ‘talk’ and skazat’ ‘say’ do occur, they show a de-
crease compared to the prediction. On the contrary, shouting verbs such as
kričat’ ‘shout’ and orat’ ‘yell’ occur with v telefon ‘into the phone’ more fre-
quently than the corpus predicts.
Table 76: Top speech verbs with the PP v telefon ‘into the phone’
Even though the top five speech verbs in Russian give us some intuitions about
the behavior of v NPACC phone PP, there is not enough data for comparison. An
example of the v NPACC PP using trubka ‘receiver’ as the filler for the NP slot
makes the data on the phone PP with preposition v more representative. The v
trubku ‘into the receiver’ PP yields twelve speech verbs. Table 77 represents the
Collostructional profiling of phone PPs in Russian and English | 189
speech verbs that occur with the v trubku ‘into the receiver’ PP more than four
times.
Table 77: Top speech verbs with the PP v trubku ‘into the receiver’
It can be seen from Table 77 that there are only eight different roots from which
the twelve speech verbs with v trubku ‘into the receiver’ are derived: burk- ‘mut-
ter’, govor- ‘talk’, krik- ‘shout’, molk- ‘remain silent’, or- ‘yell’, otvet- ‘answer’,
šept-‘whisper’ and skaz- ‘say’. Six verbs are produced using a bare verb root and
verb ending and six other verbs are derived using a prefix or a suffix. Among the
prefixes we see za- which has an ingressive meaning and is glossed as ‘start V-
ing’ (see Sokolova 2012 and references therein) and pro- which has the meaning
of producing a quantum and is glossed as ‘V something’ (see chapter 5 and
references therein). The only suffix present in the data is -nu-, which has sem-
elfactive meaning (see Makarova and Janda 2009 and references therein). How-
ever as a speech verb burknut’ has a meaning very close to the verbs prefixed in
pro- describing producing a quantum of speech: ‘mutter once’ means ‘mutter a
quantum of information’, so in section 6.4 examining the semantics of these
verbs, the verb burknut’ ‘mutter once’ is grouped with the verbs that mean ‘pro-
duce a quantum of information’.
The semantic field of speech and sound in the RNC is dominated by three
verbs: govorit’ ‘talk’, skazat’ ‘say’ and otvečat’ ‘answer’. The left pie chart of
Figure 27 shows how these verbs dominate the chart. On the contrary the verbs
||
41 Molčat’ ‘remain silent’ has a tradition of being investigated as a speech verb in Russian, see
Krongauz 1998.
190 | Collostructional profiling
that can be used with v trubku ‘into the receiver’ are dominated by a different
verb: kričat’ ‘shout’. We can see from Table 77 that kričat’ ‘shout’ occurs with v
trubku ‘into the receiver’ ten times more often than the corpus predicts. The
distribution in the corpus predicts that kričat’ v trubku ‘shout into the receiver’
should appear in six occurrences, while it is found in sixty-nine occurrences.
Fig. 27: Top speech verbs with the PP v trubku ‘into the receiver’
The two charts in Figure 27 clearly show that the distribution of the speech and
sound verbs with the PP v trubku ‘into the phone’ is different from the distribu-
tion of the same verbs in the corpus. This difference is statistically significant (χ2
= 135.35, df = 11, p-value < 2.2e-16), and the effect size is large (Cramer’s V =
0.57). Thus the first hypothesis is confirmed: these are examples of use of a
specific phone construction with specific semantic requirements on the verb
that can be used in it. The most frequent speech verbs are not frequent in this
construction and the verbs that are most frequent in this construction are not
those that are most frequent overall.
We see that both Russian and English show a tendency for shouting and
whispering speech verbs when the preposition meaning ‘into’ is involved in a
phone PP. The interesting question arises as to why these verbs are preferred by
such PPs.
This section discusses why certain verbs are repulsed from the English into the
phone/receiver and Russian v telefon/trubku ‘into the phone/receiver’, while
other verbs are attracted to these PPs. The distribution can be explained by the
nature of speaking into the phone. In this section I describe the most prominent
fillers for each language. As we recall, English into the phone has more verbs in
Collostructional profiling of phone PPs in Russian and English | 191
its collostructional profile than into the receiver (ten as opposed to four), while
in Russian v trubku ‘into the receiver’ is more prominent: it yields twelve speech
verbs as opposed to five verbs in the collostructional profile of the v telefon ‘into
the phone’.
The most neutral speech verbs in English, say and talk, show a decrease
compared to the prediction with the PP into the phone. Therefore overall the into
the phone PP repels neutral speech verbs42; see Table 78.
Among the verbs that can be used with into the phone, the most prominent
items are non-neutral speech verbs: shouting verbs (like scream, shout, yell,
bark; see Table 79) and whispering verbs (like whisper, sigh and breathe; see
Table 80).
In Russian I investigate in more detail the v trubku ‘into the receiver’ PP which
has more top speech verbs and therefore has more material for comparison. The
most neutral verbs from the list – skazat’ ‘talk’, govorit’ ‘say’, otvečat' ‘answer’
||
42 Still, it is interesting that the neutral speech verb speak, which is less frequent in the cor-
pus, occurs in this construction more frequently than the corpus predicts. The reason should
be clarified in future studies.
192 | Collostructional profiling
Table 81: Neutral verbs with the v trubku ‘into the receiver’ PP
Among the verbs that can be used with v trubku ‘into the receiver’, the most
prominent items are also non-neutral speech verbs: shouting verbs, whispering
verbs and quantization verbs. Shouting verbs are presented by verbs like kričat’
‘shout’, zakričat’ ‘start shouting’, prokričat’ ‘shout something’; and whispering
verbs in Russian are presented by only one verb: prošeptat’ ‘whisper some-
thing’; see Table 82.
Table 82: Shouting and whispering verbs with the PP v trubku ‘into the receiver’
Quantization verbs can be divided into two classes: first, verbs that mean to
‘produce a quantum of information’ (like progovorit’ ‘talk about something’ or
prošeptat’ ‘whisper something’) and second, verbs that mean ‘start speaking’
(like zakričat’ ‘start shouting’ or zagovorit’ ‘start talking’); see Table 83. Three
verbs, prokričat’ ‘shout something’, zakričat’ ‘start shouting’, and prošeptat’
‘whisper something’, belong both in the shouting and whispering class and the
quantization class, and therefore appear both in Table 82 and Table 83. We do
not obtain a class of quantization verbs in the English list of speech verbs with
into the phone, because in Russian quantization meanings are introduced by
verbal prefixes and English lacks overt mechanisms parallel to Russian prefixa-
tion.
The preference for these verbs can be explained by the nature of the situa-
tion of talking into the phone. It is important to note that the person who says a
sentence like (153) is not a participant in the communication, but the observer of
Collostructional profiling of phone PPs in Russian and English | 193
it. That person cannot be the destination point for the message said into the
phone. On the contrary, this person is located near the participant who is the
source of information in the communication into the phone and observes him or
her speaking.
Table 83: Quantization verbs with the PP v trubku ‘into the receiver’
There are three possibilities for the observer to participate in the communica-
tion. First, the speaker is talking to the listener, but the speaker is talking too
loud and the observer hears it even though he or she might be not interested in
it, as in (153). This type of situation explains the increase in frequency for the
shouting verbs, such as shout or yell. Second, the speaker might be aware of the
observer and might want to intentionally exclude the observer from the com-
munication on the phone. In this case the speaker would speak in a low voice.
This type of situation explains why whispering verbs occur frequently with PPs
meaning ‘into the phone’. Third, the speaker can participate in two communica-
tions at the same time: one with the listener on the phone and one with the
observer. In this case, in the situation of a continuing communication, channel
disambiguation is not needed. If we are talking to one addressee, we continue
talking to them. But if the communication has just started or a quantum of
communication was produced, the channel needs to be chosen, because for this
new piece of information the intended addressee is not clear. It could be intend-
ed for either the listener or the observer. Thus when we use verbs like govorit’
‘talk’ or skazat’ ‘say’, which refer to continuous communication, then we do not
need to mention whether it was into the phone or not. However if we use verbs
like progovorit’ ‘talk about something’ or zagovorit’ ‘start talking’, which denote
quantized communication, then we need to specify which channel was used for
this communication. As a result the verbs which mean ‘start talking’ or ‘say a
194 | Collostructional profiling
quantum’ occur more frequently with the v trubku ‘into the receiver’ PP which is
a way to choose the channel.
Thus, the verbs attracted to English into the phone/receiver and Russian v
telefon/trubku ‘into the phone/receiver’ are shouting verbs, whispering verbs
and quantization verbs. The preference for such verbs reflects the nature of the
situation of communicating into the phone. Summing up, it can be concluded
that English on the phone can be added almost freely to any speech verb. Rus-
sian po telefonu, v telefon/trubku and English into the phone/receiver cannot be
added freely to a neutral speech verb and produce independent constructions.
Russian po telefonu ‘on the phone’ has a preference for durative speech verbs.
‘Into the phone/receiver’ in both languages is used as an element of a phone
construction that has a preference for shouting and whispering verbs both in
Russian and English, and quantization verbs in Russian.
The case studies above describe the behavior of individual constructions. How-
ever collostructional profiling also sheds light on the behavior of the semantic
restrictions when there is an interaction of two constructions. According to the
definition of Construction Grammar, a construction is postulated if we find “a
pairing of form with meaning/use such that some aspect of the form or some
aspect of the meaning is not strictly predictable from the component parts or
from other constructions already established to exist in the language” (Goldberg
1996: 68). In more recent works this definition has been generalized to include
frequent form-meaning pairs, see more in chapter 2. As a result, most works in
the framework of Construction Grammar are devoted to describing individual
constructions. However, in discourse various constructions are frequently com-
bined and therefore understanding the principles of interaction is crucial for
future development of Construction Grammar. We come across an interaction of
two constructions when we are dealing with a pairing of form and meaning that
is largely predictable from previously analyzed constructions. The mechanisms
of interaction of form and meaning are discussed in Kuznetsova and Lyash-
evskaya 2011, while in the section below I discuss how the semantic restrictions
on a slot behave in the interaction of constructions. We will see that each con-
struction projects its preferences onto the slot in the interaction and that the
resulting restrictions present a sum of the restrictions of both constructions.
Interaction of the possessive construction and the construction of disappearance | 195
(154) My kak raz okazalis’ v Pitere, no na bal ne popali – v tot večer u Volodi
byl koncert. [Sati Spivakova. Ne vse (2002)]
‘We were in St. Petersburg on that day, but did not go to the ball - that
night Volodya had a concert.’
(156) Ona, èta ženščina, ubila Kolju Rubcova… [Viktor Astaf’ev. Zatesi
(1999) // «Novyj Mir», 2000]
‘She, this woman has killed Kolja Rubcov.’
(158) the agent previously had the headaches, but at present they have disap-
peared. In (159) the speaker had a wife in the past, but now she is not physically
present in the observable world, because she was killed.
(157) Vskore èta knižečka u menja isčezla iz zapertoj party. [P.P. Bažov.
Otsloenija dnej (1944-1947)]
‘Soon thereafter this book disappeared from my locked desk (lit. Soon
this book disappeared by me from the locked desk).’
(158) K koncu tret’ix sutok u menja isčezli golovnye boli, odyška, normal-
izovalsja son. [Jurij Senkevič. Putešestvie dlinoju v žizn’ (1999)]
‘By the end of the third day my headaches were gone, shortness of
breath disappeared, and my sleep became normal (lit. By the end of the
third day by me headaches disappeared, shortness of breath [disap-
peared], and sleep became normal’).’
The table below presents the collostructional profiling of the verbal slot in the
interaction of the possessive construction and the disappearance construction.
There are twelve verbs that appear in this slot in at least five examples of the
interaction. The first column of the table shows the verb found in the verbal slot.
The second column provides a gloss for the verb. The third column labeled
“corpus” presents how often the verb appears in the RNC. The fourth column
shows how examples would be distributed if their distribution matched the
distribution of the verbs of disappearance in the corpus. The fifth column shows
how often the verb appears in the interaction. Since this case study shows how
collostructional profiling can be combined with collostructional analysis, the
sixth column provides us with the main measure of collostructional analysis:
the probability measured via the Fisher test. This test compares the frequency of
the verb of disappearance in the corpus with the frequency of the same verb in
the interaction of the two constructions. Comparing these two numbers to the
token frequency of the verbs of disappearance in the corpus and the token fre-
quency of verbs of disappearance in the interaction, we can conclude whether
the verb of interest is attracted to or repulsed from the interaction. The column
labeled “P (Fisher test)” shows the result of the Fisher test applied to the four
values: 1) the number of the occurrences of the verb in the interaction, 2) the
number of all occurrences of a verb in the construction of disappearance ex-
Interaction of the possessive construction and the construction of disappearance | 197
Table 84: Verbs of disappearance in the interaction with the possessive construction
u NPXgen u NPXgen
Vdisappear Vdisappear
NPYnom: NPYnom: P
verb gloss corpus predicted observed (Fisher test) change
umeret’ ‘die’ 37,844 54 52 0.75
ubit’ ‘killPFV’ 34,889 50 33 0.004 ∨
končat’sja ‘end’ 5,410 7 29 1.29e-09 ∧
isčeznut’ ‘disappear’ 17,492 25 25 1
pogibnut’ ‘perish’ 17,424 25 16 0.054
prekratit’sja ‘cease’ 3,316 4 13 0.0001 ∧
issjaknut’ ‘run out’ 870 1 11 7.09e-08 ∧
skryvat’sja ‘hide’ 4,839 6 11 0.12
sžeč’ ‘burn’ 4,710 6 7 0.84
‘die (about ∨
past’ animals)’ 8,575 12 5 0.03
ubivat’ ‘killIPFV’ 8,842 12 5 0.02 ∨
TOTAL 144,211 207 207
198 | Collostructional profiling
There are three verbs that appear in the interaction of the two constructions
more often than their distribution in the corpus predicts: končat’sja ‘end’,
prekratit’sja ‘cease’ and issjaknut’ ‘run out’. It is interesting to note that all three
verbs signal the disappearance of inanimate objects.
u NPXgen u NPXgen
Vdisappear Vdisappear
NPYnom: NPYnom: P
verb gloss corpus predicted observed (Fisher test) change
končat’sja ‘end’ 5,410 7 29 1.29e-09 ∧
prekratit’sja ‘cease’ 3,316 4 13 0.0001 ∧
issjaknut’ ‘run out’ 870 1 11 7.09e-08 ∧
Let us now turn to the three verbs that appear in the corpus significantly less
frequently than their corpus distribution predicts: ubit’ ‘killPFV’, past’ ‘die (about
animals)’ and ubivat’ ‘killIPFV’. These three verbs refer to animate beings, namely
people and animals that cease to exist.
u NPXgen u NPXgen
Vdisappear Vdisappear
NPYnom: NPYnom: P
verb gloss corpus predicted observed (Fisher test) change
ubit’ ‘killPFV’ 34,889 50 33 0.004 ∨
‘die (about ∨
past’ animals)’ 8,575 12 5 0.03
ubivat’ ‘killIPFV’ 8,842 12 5 0.02 ∨
Thus preference and dispreference for verbs in the interaction provides us with
an interesting insight into the interaction of semantic restrictions of the two
constructions when these two constructions produce a hybrid construction. We
can see that the verbs associated with inanimate subjects are attracted to the
interaction. This preference is natural, because the slot for the possessed object
is more likely to have an inanimate noun than an animate noun. We can see this
preference from the distribution in the RNC. In Table 87 the patterns NP Vdisap-
pear and Vdisappear NP that together yield uses of the disappearance construc-
tion are attested with inanimate nouns in 62% of all examples, while the pattern
u NPXgen V NPYnom that yields examples of the possessive construction in 72%
of all cases is found with inanimate nouns. Thus, an inanimate noun is more
Interaction of the possessive construction and the construction of disappearance | 199
Table 87: Distribution of animate and inanimate nouns in the possessive and disappearance
constructions
Among the verbs associated with the disappearance of animate subjects, the
intransitive verbs that usually refer to people, umeret’ ‘die’, pogibnut’ ‘perish’,
are neutral in terms of the interaction. By contrast the transitive verbs that refer
to people’s disappearance, ubit’ ‘killPFV’ and its imperfective correlate ubivat’
‘killIPFV’, are repulsed from the interaction. This tendency can also be explained
by the preferences of the possessive construction: it is usually used with intran-
sitive verbs. Table 88 compares the distribution of verbs in the pattern u NPXgen
V NPYnom which yields examples of the possessive construction with the distri-
bution of verbs of disappearance in the RNC. The table demonstrates that
among the uses of the possessive construction, 75% are with intransitive verbs.
Compare this to the distribution of transitive and intransitive verbs of disap-
pearance in the RNC in the same table: 59% of all verbs in the disappearance
construction are transitive verbs. The only verb that describes the disappear-
ance of animals, past’ ‘die (about animals)’, is found in fewer examples than its
overall distribution predicts. However this is the only verb that describes the
disappearance of animals in the collostructional profile, so it does not provide
enough data to make a solid conclusion about the preferences of the interaction
as related to animal subjects.
Table 88: Distribution of transitive and intransitive verbs in the possessive and disappearance
constructions
The two constructions interact not only in their form and meaning, but also in
the semantic restrictions of the constructions. We see how verbs of disappear-
ance in the interaction inherit the semantic restrictions of the possessive con-
struction: preference for inanimate subjects and intransitive verbs. Collostruc-
tional profiling in this section was used in combination with collostructional
analysis. Collostructional analysis is used in a modified version, where instead
of the number of all constructions in the corpus we operate with all available
examples of the interaction of two constructions in the corpus. Collostructional
profiling allowed us to produce the list of the most frequent fillers of the verbal
slot. The collostructional analysis applied to the verbs in the profile demon-
strated which verbs in the interaction are attracted, repulsed or neutral. As a
result we discovered semantic restrictions that the interaction poses on the
verbs of disappearance and were able to show that these are exactly the re-
strictions that one of the constructions in the interaction has. Thus we see that
the combination of both methods allows us to investigate the slot of the con-
struction even further.
6.5. Summary
are particular to certain cultures will be filtered out in the comparison. Activities
that appear in many or all such lists will indicate assumptions about gender
roles supported worldwide.
Grammatical profiles can be applied to various parts of speech and to vari-
ous grammatical properties for which those parts of speech are inflected. We
can study case and number distribution for nouns; tense, aspect, person, and
number distribution for verbs; and gender, case, and number distribution for
adjectives. The advantage of grammatical profiles is that, like other kinds of
profiles discussed in this book, they are correlated with semantics, but unlike
most the other profiles, they are easily accessible from a corpus. Not many lan-
guages are currently represented by a corpus tagged for syntactic constructions
or submeanings of derivational affixes. However, most corpora are tagged for
morphological information. Therefore, whereas semantic profiling or construc-
tional profiling requires an extensive tagging process, the information needed
for grammatical profiling can be acquired from a morphologically tagged cor-
pus. Using grammatical profiling we can obtain a rough preliminary semantic
classification of words almost without effort. Such classification can be subse-
quently improved by using other kinds of profiles.
In chapter 4 we discover that constructions correlate with the choice of the
submeaning of a prefix used in the construction. Using material from a pilot
experiment, we see that native speakers are able to deduce the submeaning of
the prefix pri- based purely on the context. I argue that this deduction is made
in two steps. First, using a given context, the speaker determines the semantic
class of the verb used, and second, given the semantic class of the verb, the
speaker connects it with the submeaning of the prefix. A computational algo-
rithm offered in chapter 4 models the second step of this mechanism. We see
that the four submeanings of the prefix pri- are in statistical correlation with the
semantic class of the verb, and because the semantic class of the verb is deter-
mined by its context, the submeaning of the prefix is also correlated with the
context.
Semantics has long been neglected in studies of word formation. Semantic
profiling provides derivational morphologists with an empirical tool that will
help evaluate the extent to which derivational processes are influenced by se-
mantics, which may help to bring empirical approaches to derivation under the
wing of cognitive linguistics.
The second part of chapter 4 discusses the radial network of the submean-
ing of the prefix pri-, where several submeanings compete for prototype status.
Now, decades after prototype theory emerged, several case studies have
demonstrated that different features associated with the prototype can point to
different members in a radial category. We can go back to studies that describe
204 | Conclusion
a prefixed perfective. Although some researchers claim that these two types of
pairs behave differently, others believe that there is no a clear-cut boundary
that distinguishes them. The intersection rate can be used as an argument in
this debate. In further research it will be make it possible to evaluate traditional
pairs formed via a perfective and a secondary imperfective and compare the
intersection rates for these pairs with the intersection rates for pairs formed via
a simplex imperfective and prefixed perfective. Such a comparison could show
the extent to which these two types of pairs are close.
Chapter 6 investigates the relationship between the semantic restriction on
the slot of a construction and the words that fill that slot (the collostructional
profile of the slot). Using several case studies, I show that a construction poses
semantic restrictions on a slot and that the most frequent fillers of a slot all
satisfy these restrictions. For example, prepositional phrases like on the phone
and into the phone co-occur with speech verbs. On the other hand, chapter 6
shows us that we can reverse this logic and deduce the semantic restrictions
from the list of fillers. Moreover, this chapter on collostructional profiles shows
us that the elements of a construction are interrelated. For example, the prepo-
sitional phrase into the phone has a preference for shouting verbs, which can be
seen clearly from the list of verbs that co-occur with that phrase. Thus, we see
that the list of the frequent fillers of a slot correlates with the semantic re-
strictions that a construction poses on its slot.
Questions that are related to construction identification and the interaction
of constructions are important in present-day Construction Grammar. Even
though in some cases it is clear whether a certain sequence of words forms a
construction, sometimes it is less straightforward. If two constructions interact,
when do they form an independent construction and when do they produce a
compositional sum of two constructions? Collostructional profiling can help to
find answers to this question and can be used to distinguish between composi-
tional interactions and independent constructions. For the case studies present-
ed in chapter 6, collostructional profiles reflect how far the interaction in ques-
tion is from 100% compositionality. Further research into other constructions in
different languages is required in order to test whether collostructional profiling
provides a general measure of our intuitive understanding of what a construc-
tion is.
All of these case studies serve to illustrate one of the basic claims of cogni-
tive and functional linguistics, which is that form is related to meaning. Moreo-
ver, because form and meaning correlate with each other, we can use the distri-
bution of forms to predict meaning. However, the relationship between form
and meaning is not an isomorphism, viz. such relationships are better described
in terms of gradient tendencies than in terms of one-to-one correspondence. For
206 | Conclusion
||
43 In several references the name of a person was written differently in different publications
due to the fact that journals and publishing houses use different transliteration systems. In
such cases one variant of a name was used throughout the book in order to avoid confusion. In
the list of references the work cited appears with the default variant of the name and if the
variant used in the publication is different it is given after that.
References � 209
national Workshop on Treebanks and Linguistic Theories (TLT13), 2–13. Tübingen, Germa-
ny.
Briton, N. J. and Hall, J. A. 1995. Beliefs about female and male nonverbal communication. Sex
Roles 32(1-2). 79–90.
Bulygina, Tat’janaV. and Šmelev, Aleksey D. 1997. Jazykovaja konceptualizacija mira (na mate-
riale russkoj grammatiki). Moscow: Jazyki russkoj kul’tury.
Butt, Miriam. 2010. The Light Verb Jungle: Still Hacking Away.
In Amberber, Mengistu, Brett
Baker and Mark Harvey (eds.) Complex Predicates in Cross-Linguistic Perspective, 48–78.
Cambridge: Cambridge University Press.
Bybee, Joan. 2007. Frequency of use and the organization of language. New York: Oxford Uni-
versity Press.
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: Cambridge University Press.
Bybee, Joan and David Eddington. 2006. A Usage-based Approach to Spanish Verbs of 'Becom-
ing'. Language 82(2). 323–355.
Bybee, Joan and Paul Hopper (eds.) 2001. Frequency and the Emergence of Linguistic Structure.
Typological Studies in Language 45. Amsterdam: John Benjamins.
Bybee, Joan and James McClelland. 2005. Alternatives to the combinatorial paradigm of lin-
guistic theory based on domain principles of human cognition. In Nancy A. Ritter (ed.).
The role of linguistics in cognitive science. Special Issue of The Linguistic Review 22(2-4).
381–410.
Bybee, Joan and Dan I. Slobin. 2007. Rules and Schemas in the Development and Use of the
English Past Tense. In Bybee, J. Frequency of use and the organization of language, 101-
126. New York: Oxford University Press.
Cedergren, Henrietta J., and David Sankoff. 1974. Variable rules: Performance as a statistical
reflection of competence. Language 50(2). 333–355.
Cejtlin, Stella N. 2000. Jazyk i rebenok. Moscow: Vlados.
Cejtlin 2009 - Ceytlin, Stella N. 2009. Grammar Errors in Acquiring Russian As a First and
Second Language. In Markus Dace (ed.) Ad verba līberorum: Linguistics and Pedagogy
and Psychology 1(1). 4–10.
Čertkova Marina Ju., Vladimir A. Plungian, A.A. Rjabčikov, D.O. Kuznecov. 1997. Otvety na
anketu aspektologičeskogo seminara filologičeskogo fakul’teta MGU im.
M.V. Lomonosova. Voprosy jazykoznanija (3). 125–137.
Chomsky, Noam. 1957. Syntactic Structures. Janua Linguarum Series Minor 4. The Hague:
Mouton.
Chomsky, Noam. 1965. Aspect of the Theory of Syntax. Cambridge: MIT Press.
Chomsky, Noam. 1977. Essays on Form and Interpretation. Amsterdam: North Holland.
Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. New York: Prae-
ger.
Clahsen, Harald. 1999. Lexical entries and rules of language: A multidisciplinary study of
German inflection. Behavioral and Brain Sciences 22. 991–1060.
Cohen, Jacob. 1988[1977]. Statistical power analysis for the behavioral sciences (revised edi-
tion). New York: Academic Press.
Corbett, Greville. 1982. Gender in Russian: An Account of Gender Specification and Its Rela-
tionship to Declension. Russian Linguistics 6(2). 197–232.
Croft, W. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective.
Oxford: Oxford University Press.
210 � References
Dąbrowska, Ewa. 2004. Language, Mind and Brain. Some Psychological and Neurological
Constraints on Theories of Grammar. Washington DC: George Washington Universtity
Press.
Dal’, Vladimir Ivanovich. 2009. Poslovicy i pogovorki russkogo naroda. Moscow: Eksmo.
Dickey, Steven M. 2006. Aspectual pairs, goal orientation and PO-delimitatives in Russian.
Glossos 7.
Dickey, Stephen and Laura Janda. 2009. Xoxotnul, sxitril: The relationship between semelfac-
tives formed with -nu- and s- in Russian. Russian Linguistics 33(3). 229–248.
Divjak, Dagmar S. and Stefan Th. Gries. 2006. Ways of trying in Russian: clustering behavioral
profiles. Corpus Linguistics and Lingustic Theory 2(1). 23–60.
Divjak, Dagmar S. and Stefan Th. Gries (eds.) 2012. Frequency effects in language representa-
tion. Berlin & New York: Mouton de Gruyter.
Dobrushina, Mellina and Paillard 2001 – Dobrushina, Ekaterina R., Ekaterina A. Mellina and
Denis Pajar. 2001. Russkie pristavki. Mnogoznačnost’ i semantičeskoe edinstvo. Moscow:
Russkie slovari.
Doleschal, Ursula. 1997 O vzaimosvjazi grammatičeskoj kategorii roda i pola. In A. Xolod, (ed.)
Verbal’naja i neverba’naja interpretacija polovux xarakteristik, 134–155. Krivoj Rog:
Meždunarodnyj Issledovatel’skij Centr Čelovek-Jazyk-Kul’tura-Ponimanie.
Eakins, Barbara Westbrook and Eakins, R. Gene. 1978. Sex differences in human communica-
tion. Boston: Houghton Mifflin Co.
Eckert, Penelope and McConnell-Ginet, Sally. 2003. Language and Gender. New York: Cam-
bridge University Press.
Ellis, Nick C. and Fernando Ferreira-Junior. 2009. Constructions and their acquisition: Islands
and the distinctiveness of their occupancy. Annual Review of Cognitive Linguistics 7. 187–
220.
Elman, Jeffrey L. 1990. Finding structure in time. Cognitive Science 14. 179–211.
Endresen, Anna, Laura A. Janda, Julia Kuznetsova, Olga Lyashevskaya, Anastasia Makarova,
Tore Nesset and Svetlana Sokolova. 2012. Russian ‘purely aspectual’ prefixes: Not so
‘empty’ after all?. Scando-Slavica 58(2). 229–290.
Evert, Stefan. 2004. The Statistics of Word Cooccurrences: Word Pairs and Collocations. Disser-
tation, Institut für maschinelle Sprachverarbeitung, University of Stuttgart.
Evert, Stefan. 2008. Corpora and collocations. In A. Lüdeling and M. Kytö (eds.), Corpus Lin-
guistics. An International Handbook, 1212–1248. Berlin & New York: Mouton de Gruyter.
Farrar, Donald E. and Robert R Glauber. 1967. Multicollinearity in regression analysis: the
problem revisited. The Review of Economic and Statistics. 92-107.
Fillmore, Charles J. 1986.
Pragmatically Controlled Zero Anaphora. In Feder, Deborah, Mary
Niepokuj, Vassiliki Nikiforidou and Mary Van Clay (eds.) Proceedings of the Twelfth Annual
Meeting of the Berkeley Linguistics Society, 95–107. Berkeley: Berkeley Linguistics Socie-
ty.
Fillmore, Charles J. 1988. The mechanisms of ‘Construction Grammar ’. In Axmaker, Shelley,
Annie Jaisser, and Helen Singmaster (eds.) Proceedings of the Fourteenth Annual Meeting
of the Berkeley Linguistics Society, 35–55. University of California, Berkeley: Berkeley Lin-
guistics Society.
Fillmore, Charles J. 1989. Grammatical Construction Theory and the familiar dichotomies. In
Dietrich, Rainer and Carl F. Graumann (eds.) Language Processing in Social Context, 17–
38. Amsterdam: North-Holland/Elsevier.
References | 211
Fillmore, Charles J. 1999. Inversion and constructional inheritance. In Webelhuth, Gert, Jean-
Pierre Koenig and Andreas Kathol (eds.) Lexical and constructional aspects of linguistic
explanation, 113–128. Stanford, CA: CSLI.
Fillmore, Charles J. and Beryl T. Atkins. 1992. Toward a frame-based lexicon: The semantics of
RISK and its neighbors. In Lehrer, Adrienne and Eva Feder Kittay (eds.) Frames, Fields and
Contrasts, 75–102. Hillsdale, NJ: Lawrence Erlbaum Assoc.
Firth, John R. 1957. Papers in Linguistics. Oxford: Oxford University Press.
Flier, Michael S. 1975. Remarks on Russian Verbal Prefixation. Slavic and East European Jour-
nal 19(2). 218–229.
Fried, Mirjam. What is Construction Grammar? http://constructiongrammar.org/ (accessed 8
March 2015).
Fried, Mirjam and Hans C. Boas (eds.) 2005. Grammatical Constructions: Back to the Roots.
Amsterdam/ Philadelphia: Benjamins.
Fried, Mirjam and Östman, Jan-Ola (eds.). 2004. Construction Grammar in a Cross-Language
Perspective. Amsterdam: John Benjamins.
Geeraerts, Dirk. 1988. Where does prototypicality come from? In Brygida Rudzka-Ostyn (ed.)
Topics in Cognitive Linguistics, 207–229. Amsterdam/Philadelphia: John Benjamins.
Geeraerts, Dirk. 2002. The theoretical and descriptive development of lexical semantics. In
Behrens, Leila and Dietmar Zaefferer (eds.) The Lexicon in Focus. Competition and Conver-
gence in Current Lexicology, 23–42. Frankfurt am Main: Peter Lang.
Geeraerts, Dirk. 2006. Prototype Theory. In Dirk Geeraerts (ed.) Cognitive Linguistics: Basic
Readings, 141–165. Berlin: Mouton de Gruyter.
Geeraerts, Dirk. 2010. The doctor and the semantician. In Glynn, Dylan and Kerstin Fischer
(eds.) Quantitative Methods in Cognitive Semantics: Corpus-driven Approaches, 63–78.
Berlin/New York: Mouton de Gruyter.
Geeraerts, Dirk, Stefan Grondelaers and Peter Bakema. 1994. The structure of lexical variation:
Meaning, naming, and context. Berlin: Mouton de Gruyter.
Gibbs Raymond W. Jr. and Herbert L. Colston. 2006. The cognitive psychological reality of
image schemas and their transformations. In Geeraerts, Dirk (ed.) Cognitive Linguistics.
Basic Readings, 239–268. Berlin: Mouton de Gruyter.
Gilquin, Gaëtanelle. 2006. The place of prototypicality in corpus linguistics: Causation in the
hot seat. In Gries, Stefan and Anatol Stefanowitsch (eds.) Corpora in Cognitive
Linguistics: Corpus Based Approaches to syntax and Lexis, 159–192. Berlin/New York:
Mouton de Gruyter.
Glovinskaja, Marina Ja. 2001. Mnogoznačnost’ i sinonimija v vido-vremennoj sisteme russkogo
glagola. Moscow: Russkie slovari, Azbukovnik.
Glynn, Dylan. 2010. Corpus-Driven Cognitive Semantics. An introduction to the field. In Glynn,
Dylan and Kerstin Fischer (eds.) Quantitative Methods in Cognitive Semantics: Corpus-
driven Approaches, 1–42. Berlin/New York: Mouton de Gruyter.
Glynn, Dylan and Kerstin Fischer. 2010. Quantitative Methods in Cognitive Semantics: Corpus-
Driven Approaches. Berlin/New York: Mouton de Gruyter.
Glynn, Dylan and Justyna Robinson (eds.) 2014. Polysemy and synonymy: corpus methods and
applications in Cognitive Linguistics. Amsterdam/Philadelphia: John Benjamins.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Struc-
ture. Chicago: University of Chicago Press.
Goldberg, Adele E. 1996. Construction Grammar. In Brown, Keith and Jim Miller (eds.) Concise
encyclopedia of syntactic theories, 68–71. Oxford: Pergamon Press.
212 | References
Lehmann, Volkmar. 1988. Der Russische Aspect und die lexicalische Bedeutung des Verbs.
Zeitshrift für slavische Philologie, 171–194.
Lehmann, Volkmar. 1993. Die russischen Aspeckte als gestufte Kategorien (Ein Beispiel für die
Bedeutung der kognitiven Linguistik in der slavistichen Sprachwissenshaft). Die Welt der
Slaven 38(1–2). 265–297.
Levin, Beth and Malka Rappapport Hovav. 2005. Argument realization. Cambridge University
Press.
Lopatková, Markéta, Zdeněk Žabokrtský, and Karolína Skwarska. 2006. Valency lexicon of
Czech verbs: alternation-based model. In Proceedings of the fifth international conference
on Language Resources and Evaluation (LREC 2006). Genoa, Italy. http://www.lrec-
conf.org/proceedings/lrec2006/ (accessed 8 March 2015).
Lyashevskaya, Olga and Sergey Sharoff. 2010. Častotnyj slovar’ sovremennogo russkogo
jazyka (na materiale Nacional’nogo Korpusa Russkogo Jazyka). Moscow, Azbukovnik.
Makarova, Anastasia and Laura A. Janda. 2009. Do It Once: A Case Study of the Russian -ну-
Semelfactives. Scando-Slavica 55. 78–99.
Martynyuk, Alla. (1991a) Social relations and sex stereotyping in language. Papers and studies
in contrastive linguistics 26. 93–101.
Martynyuk, Alla. (1991b) A contrastive study of male and female occupational terms in English
and Russian. Papers and studies in contrastive linguistics 26. 103-110.
Maslov, Jurij S. 1984. Očerki po aspektologii. Leningrad: Leningrad State University.
Meyers, Adam. 2007. Annotation guidelines for NomBank – noun argument structure for Prop-
Bank. Technical report, New York University.
http://nlp.cs.nyu.edu/meyers/nombank/nombank-specs-2007.pdf (accessed 8 March
2015).
Mikaelian, Irina D., Alexei D. Shmelev and Anna A. Zalizniak. 2007. Imperfectivization in Rus-
sian. In Meaning-Text Theory 2007. Proceedings of the 3rd International Conference on
Meaning-Text Theory, Klagenfurt, May 20–24, 2007. Wiener Slawistischer Almanach 69.
315–323.
Mikaelian, Shmelev and Zalizniak 2010 –Mikaèljan, Irina D., Alexey D. Šmelev and Anna A.
Zaliznjak. 2010. Vidovaja korreljativnost’ v russkom jazyke: v zaščitu vidovoj pary. Vo-
prosy Jazykoznanja (1). 3–23.
Mohan, Bernard A. 1977. Acceptability testing and fuzzy grammar. In Greenbaum, Sidney (ed.)
Acceptability in Language, 133–148. The Hague: Mouton.
Mudraya, Olga, Scott S. L. Piao, Paul Rayson, Serge Sharoff, Bogdan Babych and Laura
Lofberg. 2008. Automatic extraction of translation equivalents of phrasal and light verbs
in English and Russian. In Granger, Sylviane and Fanny Meunier (eds.) Phraseology: an in-
terdisciplinary perspective, 293–309. Amsterdam: John Benjamins.
Nesset, Tore. 2001. How Pervasive are Sexist Ideologies in Grammar? In Dirven, René, Bruce
Hawkins and Esra Sandikcioglu. Language and Ideology. Volume I: Theoretical cognitive
approaches, 197–226. Amsterdam/Philadelphia: John Benjamins.
Nesset, Tore, Anna Endresen, Laura Janda, Anastasia Makarova, Francis Steen, and Mark
Turner. 2013. How ‘here’ and ‘now’ in Russian and English establish joint attention in TV
news broadcasts. Russian Linguistics, 37(3). 229–251.
Nesset, Tore, Laura A. Janda and Anna Endresen. 2011. Two ways to get out: Radial Category
Profiling and the Russian Prefixes vy- and iz-. Zeitschrift für Slawistik 56(4). 377–402.
Newman, John 2008. Aiming low in linguistics: Low-level generalizations in corpus-based
research. Keynote address presented to the 11th International Symposium on Chinese
216 | References
Ragnasdóttir, Hrafnhildur, Hanne Gram Simonsen and Kim Plunkett. 1996. Acquisition of past
tense inflection in Icelandic and Norwegian Children. In Clark, E. (ed.) Proceedings of the
28th Stanford Child Language Forum, 259–269. Stanford: CSLI.
Rakhilina, Ekaterina. 2007. Glagoly plavanija v russkom jazyke. In Majsak, Timur and Ekaterina
Rakhilina (eds.) Glagoly dviženija v vode: leksičeskaja tipologija, 267-285. Moscow: In-
drik.
Ramchand, Gillian. 2004. Time and the event: The semantics of Russian prefixes. In Svenonius,
Peter (ed.) Nord lyd, Tromsø Working Papers on Language and Linguistics 32.2: Special
issue on Slavic prefixes, 323–361. Tromsø: University of Tromsø, Tromsø.
Raskin, Viktor. 1978. On some peculiarities of the Russian lexicon. In Farkas, Donka, Wesley M.
Jakobsen and Karol W. Todrys (eds.) Papers from the Parasession on the Lexicon, 312–325.
Chicago: Chicago Linguistic Society.
Rekers, George A., Hortensia D. Amaro-Plotkin and Benson P. Low. 1977. Sex-Typed Manner-
isms in Normal Boys and Girls as a Function of Sex and Age. Child Development 48(1).
275–278.
Rice, Sally and John Newman (eds.) 2010. Empirical and Experimental Methods in Cogni-
tive/Functional Research. Conceptual Structure, Discourse, and Language Series. Stan-
ford: CSLI.
Roland, Douglas and Daniel Jurafsky.
2002. Verb sense and verb subcategorization probabili-
ties. In Stevenson, Suzanne and Paola Merlo (eds.) The Lexical Basis of Sentence Pro-
cessing: Formal, Computational, and Experimental Issues, 325–346. Amsterdam: John
Benjamins.
Romaine, Suzanne. 1999. Communicating Gender. Mahwah, NJ: Lawrence Erlbaum Associates,
Inc.
Romaine, Suzanne. 2001. A corpus-based view of gender in British and American English. In
Hellinger, Marlis and Hadumod Bußmann (eds.) Gender Across Languages. The
de/construction of gender roles through language variation and change. Volume 1, 153–
175. Amsterdam/Philadelphia: John Benjamins.
Romanova, Eugenia. 2004. Superlexical vs. lexical prefixes. In Svenonius, Peter (ed.) Nord lyd,
Tromsø Working Papers on Language and Linguistics 32.2: Special issue on Slavic
prefixes, 255–278. Tromsø: University of Tromsø.
Rosch, Eleanor.H. 1973. Natural categories. Cognitive Psychology 4 (3). 328–350.
Rosch, Eleanor. 1975. On the internal structure of perceptual and semantic categories. In
Moore, Timothy E. (ed.) Cognitive Development and Acquisition of Language, 111–144.
New York/San Francisco/London: Academic Press.
Ross, John R. 1972. The category squish: Endstation hauptwort. In Peranteau, Paul M., Judith N.
Levi and Gloria C. Phares (eds.) Papers from the 8th Meeting of the Chicago Linguistic So-
ciety, 316–338. Chicago: Chicago Linguistic Society.
Ross, John R. 1973a. A fake NP squish. In Bailey, Charles James N. and Roger W. Shuy (eds.)
New Ways of Analyzing Variation in English, 96–140. Washington: Georgetown University
Press.
Ross, John R. 1973b. Nouniness. In Fujimura, Osamu (ed.) Three Dimensions of Linguistic Theo-
ry, 137–257. Tokyo: TEC.
Rothstein, Susan. 1983. The syntactic forms of predication, Diss., MIT.
Rummelhart, David E. and James McClelland. 1986. On learning the past tense s of English
verbs: Implicit rules or parallel distributed processing? In James L. McClelland, David E.
218 | References
Rummelhart, and the PDP Research Group (eds.) Parallel distributed processing: Explora-
tions in the microstructure of cognition, 216–271. Cambridge, MA: MIT Press.
Rusakova, Marina V. and Sergej S. Saj. 2003. Vidovaja para russkogo glagola v individual’nom
leksikone i rečevoj dejatel’nosti. In Natalia V. Bogdanova and Boris I. Osipov (eds.).
Grammatičeskaja i leksičeskaja semantika. Pamjati L.L. Bulanina, 116–128. Saint Peters-
burg: Saint Petersburg State University.
Rusakova, Marina V. and Sergej S. Saj. 2008. Glagol’naja paradigma v individual’nyx sistemax
nositelej russkogo jazyka i problema grammatičeskogo vida. Vestnik SPbGU. Filologija.
Vostokovedenie. Žurnalistika 2(2). 205–214.
Rusakova et al. 2001 – Roussakova, Marina, Serguei Sai, Svetlana Bogomolova, Dmitrij,
Guerassimov, Tatiana Tangisheva and Natalia Zaika. 2001. On the mental representation
of Russian aspect relations. In Bendjaballah, Sabrina, Wolfgang U. Dressler, Oskar E.
Pfeiffer and Maria D. Voeikova (eds.) Morphology 2000: Selected papers from the 9th
Morphology meeting, Vienna, 24–28 February 2000. Current issues in linguistic theory
218, 305–331. Amsterdam/Philadelphia: John Benjamins.
Schmid, Hans-Jörg. 2000. English abstract nouns as conceptual shell. From corpus to cogni-
tion. Berlin: Mouton de Gruyter.
Schmid, Hans-Jörg. 2003. Do women and men really live in different cultures? Evidence from
the BNC. In Wilson, Andrew, Paul Rayson and Tony McEnery (eds.) Corpus linguistics by
the Lune: a Festschrift for Geoffrey Leech, 185–221. Frankfurt: Peter Lang.
Schmid, Hans-Jörg. 2007. Entrenchment, salience, and basic levels. In Geeraerts , Dirk and
Hubert Cuyckens (eds.) The Oxford Handbook of Cognitive Linguistics, 117–138. Ox-
ford/New York: Oxford University Press.
Schmid, Hans-Jörg and Susanne Handl (eds.) 2010. Cognitive foundations of linguistic usage
patterns. Empirical approaches. Berlin etc.: Mouton de Gruyter.
Schmid, Hans- Jörg, and Helmut Küchenhoff. 2013. Collostructional analysis and other ways of
measuring lexicogrammatical attraction: Theoretical premises, practical problems and
cognitive underpinnings. Cognitive linguistics 24(3). 531–577.
Schmid, Sonja. 1998. Zur Bezeichnung weiblicher Personen im Russischen: Eine empirische
Pilotstudie. Wiener Slawistischer Almanach 41. 239–262.
Schooneveld, Cornelius H. van. 1958. The so-called ‘prèverbe vides’ and neutralization. Dutch
Contributions to the Fourth International Congress of Slavistics, 159–161. The Hague: Mou-
ton.
Simonsen, Hanne Gram. 2000. Past tense acquisition and processing in Norwegian: Experi-
mental evidence. Language and Language Behavior 3(2). 86–101.
Sinclair, John. 1991. Corpus, concordance, collocation: Describing English language. Oxford:
Oxford University Press.
Smith, Philip M. 1985. Language, the Sexes and Society. Oxford, New York: Basil Blackwell.
Sokolova, Svetlana. 2012. Asymmetries in Linguistic Construal: Russian Prefixes and the Loca-
tive Alternation. Diss., University of Tromsø.
Sokolova, Svetlana, Olga Lyashevskaya and Laura A. Janda. 2012. The Locative Alternation and
the Russian ‘empty’ prefixes: A case study of the verb gruzit’ ‘load’. In Divjak, Dagmar and
Stefan Th. Gries (eds.) Frequency Effects in Language Representation. Volume 2, 51-86.
Berlin: Mouton de Gruyter.
Steels, Luc. (ed.) 2011. Design Patterns in Fluid Construction Grammar. Amsterdam: John Ben-
jamins.
References | 219
Steels, Luc. (ed.) 2012. Experiments in Cultural Language Evolution. Amsterdam: John Benja-
mins.
Stefanowitsch, Anatol and Stefan Th. Gries. 2003. Collostructions: investigating the interac-
tion between words and constructions. International Journal of Corpus Linguistics 8(2).
209–243.
Stefanowitsch, Anatol, and Gries, Stefan Th. 2005. Covarying collexemes. Corpus Linguistics
and Linguistic Theory 1(1). 1–43.
Stefanowitsch, Anatol and Stefan Th. Gries (eds.) 2006. Corpus-based approaches to meta-
phor and metonymy. Berlin/New York: Mouton de Gruyter.
Stubbs, Michael. 1995. Collocations and semantic profiles: on the cause of the trouble with
quantitative studies. Functions of Language 2(1). 23–55. [Reprinted in Teubert, Wolfgang
and Ramesh Krishnamurthy (eds.) 2007. Corpus Linguistics: Critical Concepts in Linguis-
tics, Volume 3, 166–193. London: Routledge.]
Švedova, Natalija Ju. et al. 1980. Russkaja grammatika, Volume I. Moscow: Nauka.
Svenonius, Peter. 2004a. Slavic prefixes inside and outside VP. In Svenonius, Peter. Nord lyd,
Tromsø Working Papers on Language and Linguistics 32.2: Special issue on Slavic
prefixes, 255–278. Tromsø: University of Tromsø.
Svenonius, Peter (ed.) 2004b. Tromsø Working Papers on Language and Linguistics: Nord lyd
32.2: Special issue on Slavic prefixes. Tromsø: University of Tromsø.
Szmrecsanyi, Benedikt. 2006. Morphosyntactic persistence in spoken English. A corpus study
at the intersection of variationist sociolinguistics, psycholinguistics, and discourse analy-
sis. Berlin & New York: Mouton de Gruyter.
Tannen, Deborah. 1990. You Just Don't Understand: Women and Men in Conversation.
Timberlake, Alan. 2004. A Reference Grammar of Russian. Cambridge: Cambridge University
Press.
Tomasello, Michael. 2003. Constructing a language: a usage based theory of language acqui-
sition. Cambridge: Harvard University Press.
Tsohatzidis, Savas L. 1990. Introduction. In Savas L. Tsohatzidis (ed.) Meanings and Proto-
types. Studies in Linguistic Categorization: Routledge, 1–13. London/New York.
Townsend, Charles E. 1968[1975]. Russian word-formation. Cambridge, MA: Slavica Publishers.
Trubetzkoy, Nikolai S. 2001. Studies in General Linguistics and Language Structure. Duke
University Press.
Vey, Marc. 1952. Les prèverbes ‘vides’ en tchèque moderne. Revue des ètudes slaves 29. 82–
107.
Wierzbicka, Anna. 1988. The semantics of grammar. Amsterdam: Benjamins.
Yokoyama, Olga T. 1986. Lexical frequency and its implications: the case of contemporary
edited Russian. Slavic and East European Journal 30(2). 147–166.
Yokoyama, Olga T. 1999. Russian genderlects and referential expressions. Language in Society
28. 401–429.
Zalizniak 1980 – Zaliznjak, Andrej A. 1980. Grammatičeskij slovar' russkogo jazyka. Moscow:
Russkij jazyk.
Zalizniak and Mikaelian 2014 – Anna A. Zaliznjak & Mikaèljan, Irina D. 2014. Russkaja glag-
ol’naja prefiksacija i problema vidovoj parnosti. Mundo Eslavo 13. 19–33.
Zalizniak and Shmelev 1997 – Zaliznjak, Anna A. and Aleksey D. Šmelev. 1997. Lekcii po russ-
koj aspektologii. München: Verlag Otto Sagner.
Zalizniak and Shmelev 2000 – Zaliznjak, Anna A. and Aleksey D. Šmelev. 2000. Vvedenie v
russkuju aspektologiju. Moscow: Jazyki russkoj kul’tury.
220 | References
Zasorina, Lidija N. (ed.) 1977. Častotnyj slovar' russkogo jazyka. Moscow: Russkij jazyk.
Appendix
Intersection rates for seventeen pairs of verbs: imperfectives and perfectives produced via the
prefix pro-
This appendix contains the constructional profiles of the seventeen pairs of verbs, and the
intersection rate calculated for each pair. All examples come from the FrameBank database.
kontrolirovat’/prokontrolirovat’ ‘control’
Table 89: Constructional profile of the verb kontrolirovat’ ‘control’ (4 constructions, 2 frequent
constructions)
Table 90: Constructional profile of the verb prokontrolirovat’ ‘control’ (4 constructions, 3 fre-
quent constructions)
Table 91: Intersection of constructional profiles of the verbs kontrolirovat’ and prokontroliro-
vat’ ‘control’
dlit’sja/prodlit’sja ‘last’
Table 92: Constructional profile of the verb dlit’sja ‘last’ (9 constructions, 4 frequent construc-
tions)
Table 93: Constructional profile of the verb prodlit’sja ‘last’ (10 constructions, 3 frequent con-
structions)
Table 94: Intersection of constructional profiles of the verbs dlit’sja and prodlit’sja ‘last’
demonstrirovat’/prodemonsrirovat’ ‘demonstrate’
Table 95: Constructional profile of the verb demonstrirovat’ ‘demonstrate (11 constructions, 4
frequent constructions)
Table 97: Intersection of constructional profiles of the verbs demonstrirovat’ and prodemon-
strirovat’ ‘demonstrate’
Table 98: Constructional profile of the verb golosovat’ ‘vote; hitch a ride’ (14 constructions, 3
frequent constructions)
Table 99: Constructional profile of the verb progolosovat’ ‘vote; hitch a ride’ (9 constructions,
3 frequent constructions)
Table 100: Intersection of constructional profiles of the verbs golosovat’ and progolosovat’
‘vote; hitch a ride’
informirovat’/proinformirovat’ ‘inform’
Table 101: Constructional profile of the verb informirovat’ ‘inform’ (7 constructions, 5 frequent
constructions)
Table 102: Constructional profile of the verb proinformirovat’ ‘inform’ (6 constructions, 4 fre-
quent constructions)
Table 103: Intersection of constructional profiles of the verbs informirovat’ and proinformiro-
vat’ ‘inform’
Table 104: Constructional profile of the verb čitat’ ‘read; recite’ (14 constructions, 4 frequent
constructions)
Table 105: Constructional profile of the verb pročitat’ ‘read, recite’ (15 constructions, 3 frequent
constructions)
Table 106: Intersection of constructional profiles of the verbs čitat’ and pročitat’ ‘read, recite’
citirovat’/procitirovat’ ‘cite’
Table 107: Constructional profile of the verb citirovat’ ‘cite’ (10 constructions, 4 frequent con-
structions)
Table 108: Constructional profile of the verb procitirovat’ ‘cite’ (6 constructions, 4 frequent
constructions)
Table 109: Intersection of constructional profiles of the verbs citirovat’ and procitirovat’ ‘cite’
kričat’/prokričat’ ‘shout’
Table 110: Constructional profile of the verb kričat’ ‘shout’ (12 constructions, 2 frequent con-
structions)
Table 111: Constructional profile of the verb prokričat’ ‘shout’ (10 constructions, 3 frequent
constructions)
Table 112: Intersection of constructional profiles of the verbs kričat’ and prokričat’ ‘shout’
šeptat’/prošeptat’ ‘whisper’
Table 113: Constructional profile of the verb šeptat’ ‘whisper’ (8 constructions, 5 frequent
constructions)
Table 114: Constructional profile of the verb prošeptat’ ‘whisper’ (5 constructions, 3 frequent
constructions)
Table 115: Intersection of constructional profiles of the verbs šeptat’ and prošeptat’ ‘whisper’
zvučat’/prozvučat’ ‘sound’
Table 116: Constructional profile of the verb zvučat’ ‘sound’ (8 constructions, 5 frequent con-
structions)
Table 117: Constructional profile of the verb prozvučat’ ‘sound’ (12 constructions, 5 frequent
constructions)
Table 118: Intersection of constructional profiles of the verbs zvučat’ and prozvučat’ ‘sound’
gremet’/progremet’ ‘resound’
Table 119: Constructional profile of the verb gremet’ ‘resound’ (11 constructions, 4 frequent
constructions)
Table 120: Constructional profile of the verb progremet’ ‘resound’ (10 constructions, 5 frequent
constructions)
Table 121: Intersection of constructional profiles of the verbs gremet’ and progremet’ ‘resound’
Table 122: Constructional profile of the verb zvenet’ ‘ring, clank’ (13 constructions, 3 frequent
constructions)
Table 123: Constructional profile of the verb prozvenet’ ‘ring, clank’ (8 constructions, 3 fre-
quent constructions)
Table 124: Intersection of constructional profiles of the verbs zvenet’ and prozvenet’ ‘ring,
clank’
pet’/propet’ ‘sing’
Table 125: Constructional profile of the verb pet’ ‘sing’ (15 constructions, 5 frequent construc-
tions)
Table 126: Constructional profile of the verb propet’ ‘sing’ (12 constructions, 2 frequent con-
structions)
Table 127: Intersection of constructional profiles of the verbs pet’ and propet’ ‘sing’
gudet’/progudet’ ‘buzz’
Table 128: Constructional profile of the verb gudet’ ‘buzz’ (13 constructions, 5 frequent con-
structions)
Table 129: Constructional profile of the verb progudet’ ‘buzz’ (8 constructions, 3 frequent
constructions)
Table 130: Intersection of constructional profiles of the verbs gudet’ and progudet’ ‘buzz’
vesti/provesti ‘lead’
Table 131: Constructional profile of the verb vesti ‘lead’ (19 constructions, 3 frequent construc-
tions)
Table 132: Constructional profile of the verb provesti ‘lead’ (12 constructions, 5 frequent con-
structions)
Table 133: Intersection of constructional profiles of the verbs vesti and provesti ‘lead’
tjanut’/protjanut’ ‘pull’
Table 134: Constructional profile of the verb tjanut’ ‘pull’ (43 constructions, 4 frequent con-
structions)
Table 135: Constructional profile of the verb protjanut’ ‘pull’ (15 constructions, 3 frequent
constructions)
Table 136: Intersection of constructional profiles of the verbs tjanut’ and protjanut’ ‘pull’
bit’/probit’ ‘beat’
Table 137: Constructional profile of the verb bit’ ‘beat’ (38 constructions, 5 frequent construc-
tions)
Table 138: Constructional profile of the verb probit’ ‘beat’ (21 constructions, 5 frequent con-
structions)
Table 139: Intersection of constructional profiles of the verbs bit’ and probit’ ‘beat’