Professional Documents
Culture Documents
Eorge Oddard: Felt Some Trees Felled Some Trees
Eorge Oddard: Felt Some Trees Felled Some Trees
FOLLOWING A BRIEF EXPLANATION of –t/d deletion and why it may occur, an account of its
phonological context, but it will also be shown that other factors may well have an
effect. Hudson (1997:3) for example believes that, although there is phonetic
motivation, Gregory Guy’s statistical analysis (1980 to 1993) has shown ‘beyond
reasonable doubt’ that this feature’s variation is ‘not a simple phonetic matter’. Since,
as Bayley (1994:304) puts it, -t/d deletion is ‘one of the most extensively studied
variable processes in English’, it will not be possible to address every angle from which
An elementary explanation of this feature is the coronal stops /t/ and /d/ not
segment is known as deletion and represents the final stage of lenition, the process of
either present or it is not, which excludes any notion of hierarchy or gradation, such as
glottalisation. Although Bybee (2002) argues that ‘a final /t/ or /d/ may vary in length’
and that ‘reduction ... can be gradual’ (p265). There are some complexities though
which effect –t/d deletion or the possibility of analysing it. For example, the preceding
consonant: if /t,d/ follow a nasal /n,m/, this is only favourable for /d/ deletion, while
/t/ will tend to remain with an altered place of articulation, generally being produced as
[ʔ]. This debuccalisation probably happens after /l/ also, as in [fɛlt sʌm tɹiːz] → [fɛlʔ sʌm
tɹiːz] (felt some trees) versus [fɛld sʌm tɹiːz] → [fɛl sʌm tɹiːz] (felled some trees). Also
/ɹ/: in a non-rhotic variety, /t,d/ will not follow /ɹ/ word finally, but where there is
rhoticity a word such as fort may be produced as [fɔɹt], and, while deletion may occur
here, it is difficult to hear owing largely to /ɹ/ and /t,d/ being homorganic. The
following consonant can produce difficulty likewise by also being /t,d/, since [dʒəst] or
[dʒəs] preceding [teɪk] will both sound very similar to [dʒəsteɪk] (just take), for example.
Wolfram (1969: 48) highlighted this ‘neutralising’ problem, discounting ‘clusters which
1 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
are immediately followed by a homorganic stop’. Temple (2009) looked at such effects,
gesture by surrounding consonants, in natural speech data, and raising the question of
whether /t,d/ not being heard in these contexts is enough to categorise them as
deletion. Leading on from this, the fact that –t/d deletion study is modelled primarily on
production rather than perception is highlighted and its status as a phonological rule at
all is challenged: ‘there is no phonological contrast between /t,d/ and zero’ (p168). It
could be said, therefore, that exactly what –t/d deletion is in terms of linguistic study is
tendency to reduce consonant clusters’ while Spencer (1996:64) refers to this ‘common
strategy’ as truncation and cites some examples from Russian where –t/d deletion
the ‘constant conflict between the principle of minimum effort and the demands of
a force for change (p45). It would seem then that if a speaker can get away with
deleting –t/d, that is, if the hearer can comprehend their speech without it, the fact that
doing so requires less effort is a real motivation. To the extent, in fact, that t/d-
deletion is ‘universally variable’, according to Guy and Boberg (1997). They assert that
‘every speaker that has been observed deletes some, but not all, of their final stops’
(p151). The analytical question therefore is not whether or not it happens but why the
FOCUSING ON THE LINGUISTIC CONSTRAINTS, Labov (1989:90) brings together the work up to
different authors. These will now be taken individually and discussed in light of recent
findings.
2 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
Firstly, the effect of the following phonological segment. Since –t/d deletion only occurs
word-finally, this constraint relates to the onset of the following word, or a pause.
Bayley (1994:304-305) puts it that speakers of all dialects in which –t/d deletion has
been studied are ‘more likely to delete/t,d/ when it is followed by a vowel than when it
from all other studies, including his own, the functional explanation of cluster
simplification and the notion of saliency (that is, /t,d/ before a vowel is more likely to
be heard to due to the contrast in sonority and the loss will therefore be more
noticeable, presumably leading to less deletion). His figures show 70% deletion before
obstruents, the highest rate, versus 25% before vowels, the lowest rate (p310). The
Liquids
Glides
Vowels
It could be concluded then that Bayley was simply mistaken is his statement. The
situation is not so clear-cut, however. Bybee (2002) considers –t/d deletion in relation
to the probability of /t,d/ occurring in certain phonological contexts, arguing that there
the following segment, she argues it is less uniformly present than the preceding
(word-internal) environment, discussed below, for a particular item and therefore has
less effect: ‘for example, the t in words ending in –nt actually deletes more often before
vowels than consonants’ (p274). The preceding n was found to condition more
consistently than the preceding segment in Guy (1991). This finding is not universally
3 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
accepted, however; in fact, Temple (2009) summaries that in ‘most if not all studies’ of
–t/d deletion (including hers), following segment has been found to have the strongest
effect (p147). This seems to cast doubt on that aspect of Bybee’s argument.
the probability of a following vowel is greater, such as: verbs occurring before vowel-
initial particles and prepositions (picked on, pulled away, pulled out, and so on), before
vowel-initial pronouns ([ɪt, ʌs, ɪm, ɜ(ː/ɹ), ɛm] (it, us, him, her, them), and, commonly
used phrases (looked alike), or the indefinite article, the sequence may become stored
and processed together as ‘conventionalised units’ (p279), and lead to more deletion
The two sequences in (2) are identical in relation to every constraint except
following segment, most studies suggest that if a speaker had never heard or spoken
them before, they would be more likely to delete (a) than (b) since their reduction would
not be as salient while making the production easier. After using these phrases for a
period of time, however, it may be that if (b) was used more often, which seems likely
due to the relative frequency of the words bout and out, the chance of deletion would
increase since it would become stored as a single unit while (a) would be re-constructed
every time.
Bybee (2002:264) presents figures from a previous study which show 45.6% retention in
high frequency words (occurring 35 times or more per million) versus 65.7% where the
frequency was lower. She cite results showing similar or even more substantial
differentiation from two other, independent studies. This effect, then, seems very well
established.
4 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
The second constraint to be accounted for has already been mentioned: preceding
consonant since, as Guy and Boberg (1997:155) list it in their Table 2, preceding vowels
consonant’s effect seems to depend on the nature of the analysis. Labov (1989:90), for
example, lists it as a ‘relatively weak constraint’ and Guy and Boberg (1997:153)
manner of articulation, with differentiation also being drawn between sibilant and non-
sibilant fricatives, and between lateral consonants and /ɹ/ for liquids, since this seemed
Stops
Nasals
Non-sibilant fricatives
According to (3) therefore [kist] (kissed) will generally be produced without /t/ more
than [kilt] (kilt). This seems to be indisputable cross-dialectally, but the distinctions
between some of the intermediate groupings are far more tenuous. Guy and Boberg’s
factor weight of the non-sibilant fricatives is 0.66, for example, while for laterals it is
0.69, giving a difference of 0.03 (ibid). To say that [mift] (miffed) is more likely to delete
/t/ than [mist] (mist), then, is almost insupportable according the those data. Indeed
the data in Temple (2009:146), taken from an earlier study, produces a rather dissimilar
ranking: /s/ > other sibilant > nasal > /l/ > stop > other fricative. It clear from these
5 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
analyses that preceding consonant does have an effect, but the exact nature of that
The innovation of Guy and Boberg (1997) was to apply the Obligatory Contour
Principle (OCP) to this problem and the resultant statistics seem far more convincing. In
the strictest sense, the principle states the prohibition of elements occurring adjacently
which are identical. Guy and Boberg water this down somewhat, taking the notion that
identical elements together are disfavoured and therefore less likely to be produced.
(4) shows /s/ sharing two features with /t/ while /l/ shares only one, which also
provides an explanation for the kissed/kilt example. The former is more likely to delete
the /t/ than /l/ because they have more shared features, which the OCP states is
prohibited, or undesirable. One aspect which illustrates this well is the nasal group:
while /n/ shares two features [+con, -cont] with /t,d/, /m,ŋ/ share only one [-cont],
and separating them produces a ‘big difference’ as Guy and Boberg’s proposal
predicted. The percentage of deletion for /m,ŋ/ was 11 while for [n] it was 46, a
difference of 35%. It could be argued that this points directly to superiority of the OCP
explanation over manner of articulation, which groups nasals together and produces
variable results.
and the place of articulation of the preceding segment, providing data for deletion
being more likely following a ‘homorganic alveolar’: 29.5% versus 21.0% (p50). This
6 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
Against this, examples such as just deleting more than bust would imply that the
OCP explanation is either incorrect (since the phonological features of the preceding
segment in this pair are identical) or is only part of the picture. It may be that deletion
is effected by both the OCP and other factors simultaneously. Bybee’s (2002) analysis,
for example, provides a model which involves every phonetic variant of a word which is
experienced being stored in a cluster of exemplars, and those which are repeated more
often grow stronger while the less frequent ones fade. The model posits that the more
frequent a word, the greater the chance that the reduced exemplar will be chosen in a
given context, thus advancing the reductive change. Since just then is more frequent, as
mentioned by Labov (1989), than bust, the exemplars which have undergone reduction
Temple (2009: 151) also adds complexity to the preceding segment constraint
preceding segment leads to unevenness in the figures for her data, which suggests an
interaction between the two, which could be argued to lessen the significance of each.
This constraint has been phrased in various ways, but it is generally agreed that there
are three morphosyntactic groupings for –t/d deletion contexts: monomorphemic, and
bimorphemic, which is subcategorised into regular and non-regular past tense verbs.
The tokens in the last group are often referred to as ‘semi-weak’ verbs and represent
more distributional complexity than the other two categories. Disregarding this for now,
most works1 agree that /t,d/ is less likely to be deleted if it is functioning as a regular
past tense or participle suffix than if it appears as part of a stem’s makeup. There is a
straightforward functional explanation of this: in mist, /t/ has the same function as
/m,i,s/, in that it distinguishes the mist stem from any other stem, for example most,
1
Neu (1980) provides results to this effect and cites 6 other works with which her findings
are ‘consistent’ (p44).
7 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
whilst in missed, /t/ is a separate morpheme with the function of distinguishing this
verb form from its root lexeme MISS. It carries more information then that the /t/ of
mist. Also mist and most, for example, differ in their vocalic element, [i] ~ [oʊ] which
carries greater sonority and phonological contrast than [t] ~ [Ø]. Consequently, the /t/
of missed is more likely to be retained. In Guy and Boyd (1990), for example, the
figures for probability of deletion in uninflected tokens is .65 versus .31 for regular past
tense verbs (p7). This statistically significant differentiation seems to apply cross-
For semi-weak tokens, however, this is not the case. There is a distributional
pattern correlating to the social variable of age: Guy and Boyd’s paper demonstrates
that ‘the probability of deletion declines with increasing age’ (ibid). The explanation
given is that the separate categorisation of semi-weak verbs takes a minimum period of
time to develop. Children group them with strong verbs which ablaut like semi-weak
verbs but do not add /t,d/, such as meet, which has the past form met [miːt → mɛt].
This means there is no underlying /t,d/ until the semi-weak group develops, leading to
stored pairs such as [liːv → lɛf] for leave → left. Any production of /t,d/ is argued to be
the result of borrowing from speakers for whom it is underlying, and during the
transition into the mature stage, a speaker is argued to view the /t,d/ as irregular
Despite the this complexity, the morphological property constraint has been
argued to have regular distribution to the extent that it can be predicted fairly
accurately on the basis of a mathematical model. Guy (1991) proposed the Exponential
Model which Hudson (1997:3) argues, from the literature up until that point, there is no
reason to doubt as a ‘real phenomenon’. Put simply, it states that the probability of
-t/d deletion in semi-weak verbs is that of regular past tense verbs squared, while in
the monomorphemic group it is the same cubed. Guy (1994) demonstrates that
predictions using this theory are correct to within 1% in all but the semi-weak category
8 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
for three separate corpora. While one did not include semi-weak verbs, for the other
two the predictions for this group were within 3% and 6.2% respectively. Although these
latter figures are less convincing, Guy (1991:10) cites low token numbers and age
variability as possible reasons. The Exponential Model then seems to be borne out
acknowledging that the Model has ‘become generally accepted’, and despite the trend
identified being in the direction it would predict, did not select the morphological
constraint as significant (p148). While for the other two constraints analysed, following
and preceding segment, the range of weight factors was 65 (.84 - .20) and 41 (.68 - .
27) respectively, for morphological class it was a 8 (.53 - .45), which is relatively
insignificant. These figures are taken from Table 1 (p146). Although ultimately it is the
theoretical explanation proposed by Guy, as discussed below, and the whether –t/d
Temple (2009:167), these data do cast some doubt on the Exponential Model. This
the American studies (p145). Also, it does seem to contradict many other studies and
(1991) explanation of the Model, which uses Lexical Phonology to account for the
applied three times to the monomorphemic group, twice to the semi-weak group and
only once to the regularly inflected group would explain the Model. Lexical Phonology
proposed such an infrastructure with all lexical items having to pass through at least
three levels before being realised, two within the lexicon and one post-lexically which
9 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
take into account the context of the production outside of the word boundaries. In
acquire underlying /t,d/, at level 2, while a semi-weak form undergoes it at level 1, and
a monomorphemic form has /t,d/ from the earliest stage of derivation (p7). Guy goes
on to argue that since –t/d deletion has been shown to be affected by the following
segment and morphological structure, the rule which deletes it could be applied at any
monomorphemic form three times, a semi-weak form twice and a regularly inflected
form just once. This analysis is not accepted by Hudson (1997), Temple (2009) and
Bybee (2002).
Hudson accepts the Exponential Model but not the Lexical Phonology
explanation, while Temple disagrees with both based on her data. Hudson (1997) does
not question the internal structure of Guy’s argument but objects to some of the
underlying assumptions, such as the ability of the deletion rule to apply on any level,
while Lexical Phonology assigns rules to specific levels (p4). He also believes Guy’s
exempting of –t/d deletion from the Strict Cycle Condition on the basis that it is a
variable rule and not a categorical one requires there to be a formal distinction between
the two, which he seems to doubt (p4), while Temple (2009) argues that it has not been
possible firstly to show that the distribution of –t/d deletion variability is consistent
the need to justify a more abstract phonological analysis’, such as it being a Continuous
Speech Process which varies and has some patterns which can be aligned with other
English CSPs (p168). This may include other linguistic constraints which have not yet
been mentioned such as the number of preceding consonants, where the more there
are, the more deletion is favoured due to ease of production; syllabic stress, where
10 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
since they are less likely to be ‘missed’ by the hearer; and voicing agreement, where the
reference to frequency effects, however. Bybee (2002) also contends Guy’s Lexical
that –t/d deletion can best be accounted for primarily as a variable Continuous Speech
Process without a unifying abstract explanation, but one which does seem to pattern
according to the OCP to some extent and is effected by the frequency of each word’s
contexts. In short, -t/d deletion variability is the result of interaction between these
factors leading to patterns but not without exceptions or complexities. The Exponential
Model, for example, seems dubious in light of the data presented in Temple (2009).
2
Word count: 3,400
11 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION
GEORGE GODDARD LG 405
REFERENCES
Bayley, Robert James. 1994. "Consonant cluster reduction in Tejano English." Language
Variation and Change 6(3): 303-326.
Bybee, Joan. 2002. “Word frequency and context of use in lexical diffusion of
phonetically conditioned sound change.” Language Variation & Change 14(3): 261-290.
Guy, Gregory and Charles Boberg. 1997. "Inherent variability and the obligatory contour
principle." Language Variation and Change 9(2): 149-164.
Guy, Gregory & Sally Boyd. 1990. "The development of a morphological class."
Language Variation & Change 2(1): 1-18.
Hudson, Richard. 1997. “Inherent variability and linguistic theory.” Cognitive Linguistics
8: 73-108.
Labov, William. 1989. "The child as linguistic historian." Language Variation & Change
1(1): 85-97.
Temple, Rosalind. 2009. “(t,d): the variable status of a variable rule.” Oxford University
Working Papers in Linguistics, Philology & Phonetics Vol 12.
12 ACCOUNTING FOR –t/d DELETION AND THE LINGUISTIC CONSTRAINTS ON ITS VARIABLE
DISTRIBUTION