Professional Documents
Culture Documents
Local Wellsdeaccenting Struct Eng Intonation PDF
Local Wellsdeaccenting Struct Eng Intonation PDF
Local Wellsdeaccenting Struct Eng Intonation PDF
intonation 1
Abstract
Introduction
not seem to have been explored in any depth. In fact, the issue has
traditionally been regarded as being within the purview of syntax and
semantics, rather than phonology.
Once it is accepted that semantic information unmediated by syntactic
structure is centrally involved in 'sentence stress' location, the principle of
not multiplying explanations unnecessarily would seem to require that
other aspects of 'sentence stress' location should as far as possible be
explained in terms of the same principles, i.e. of information focus and
structure: the distribution of 'given', 'new', and 'contrastive' information.
In the present paper it will be shown that a recent excursion into this field by
generative linguists (Zwicky and Levin 1980) fails to arrive at a satisfactory
explanation of the data in question precisely because they do not adhere to
Occam's principle, preferring instead to account for apparent peculiarities
in the location of 'contrastive stress' in purely syntactic terms.
Although there are many difficulties involved in making exponency
statements for the phonological category 'sentence stress', and indeed in
admitting such a category at all, we comply, in the present paper, with the
traditional generative approach to sentence stress by eschewing reference
to the phonetic exponents of proposed phonological categories, except
where we find it quite unavoidable. We do this to demonstrate that
Zwicky and Levin's approach can be refuted within its own terms. In
order to emphasize that the discussion is at the phonological level and
implies prior abstraction from the phonetic data, we prefer the term
'sentence accent', following Bolinger (1972) and Ladd (1980), since the
term 'stress' has unfortunately been used both at the phonetic and at the
phonological level. A further difficulty with the term 'sentence stress' is its
claim that the domain of this accentual phenomenon is the sentence. On the
basis of recent empirical work (Local et al. forthcoming) we have proposed
that a more appropriate domain is the speaker's turn. For the purposes of
this analysis, however, we retain the term 'sentence', to emphasize that the
examples we cite are not drawn from recordings of naturally occurring
conversation.
In their squib 'You don't have to\ Zwicky and Levin (1980; hereafter Z.
and L.) identify a class of items that cannot receive 'contrastive' stress.
These are 'non-finite verbals (infinitoid or participle) followed by a gap
created by VP deletion'. The class of 'infinitoid' comprises infinitival to,
together with have and be as nonfinite forms. They conclude their article
with a series of open questions, to which we shall suggest some answers:
Why should this bei Why should this deletion rule be the offender, while other
rules are not? Why should zero anaphora offend, when anaphora with do so ...
does not. ... What is there about contrastive stress that calls up the constraint?
And finally, why should only non-finite verbals be affected?
make two observations. First, new information can be conveyed not only
by introducing a new lexical item but also by less obvious means, such as
changing negative to affirmative, as in (1) to (3); and/or changing the tense
or modality of or otherwise modifying the verb. In such cases, there may
not always be a semantically accentable item to bear the accent and thus
mark the new information function. Second, Ladd (1980) points out that
the accent has to go somewhere. If there is no suitable location for it, in
terms of the distribution of new and given information, it will end up on
some word by default: otherwise the sentence would be interpreted as
incomplete. Thus in the second part of (Ã), in line with our first
observation, the speaker might present as the main point of new
information either the contrast in time reference, with the accent on this
time, or else, as Z. and L. imagine, modification of the verbal constitu-
ent. In the latter case, the candidates for accent location are very
restricted: I'm is ruled out, because this would present / as new again,
when in fact it is given; the same goes for see and it. This leaves likely and
to as candidates (though note that if the uncontracted form / am were
used, am would be a strong candidate too). We shall show that what Z.
and L. present as a syntactic problem can be resolved by pursuing the
implications of these two observations.
What is at issue here is not the movement rule, but the previous
occurrence of the lexical item have. The same observation holds for (5)
and (5') (we ring the relevant antecedents):
(5) At first you said you were going into the marines, now you're
planning to go to theological school. What do you want to be when
you grow up?
(5') ?? At first you said you were going to be & priest; now you're planning
to(ße)a marine. What do you want to be when you grow up?
If the lexical item be has already occurred, it is difficult to accent it in the
second sentence. Z. and L.'s examples of topicalization and though
inversion are susceptible to parallel counterexamples. We list Z. and L.'s
sentences followed by our own, in which lexical repetition reduces
acceptability quite considerably.
(6) I don't buy many clothes, but that suit I want to have.
(6') ? I don't (have) many clothes, but that suit I want to have.
(7) I'm a competent researcher, but a good teacher I could never be,
(T) ? I may (6e) a competent researcher, but a good teacher I could never
be.
(8) Quiet though he tried to be, Alex made enough noise to wake the
dead.
(8') ? Alex was bound to @ noisy, quiet though he tried to be.5
(9), which is Z. and L.'s example of right-node raising, seems to be of a
quite different order:
(9) Barb decided to — but Leo decided not to — participate in the
discussion.
Although there is a syntactic gap after Barb decided to, this gap cannot
coincide with sentence-final position, as was the case for (6) to (8).
Another interesting feature of (9), probably a consequence of this, is that
the possible types of accent on to are highly restricted: it must be either a
rise or a fall-rise (pitch movements whose functions include projecting
that more talk will follow).
However, such an accent can also be used before a gap created by VP
deletion (not normally possible) provided the 'infinitoid' is not sentence-
final. Thus (10) to (12), asterisked by Z. and L., are acceptable if the first
accent is a fall-rise with a large amount of pitch movement and if there is a
second accent later in the sentence:
(10) I don't think you'll have anyone want to take early exams. For
anyone to ask to is most un/zfcely.
(11) Tony would have preferred not to sleep on a wooden pillow.
However, he's been persuaded to by his friends.
(12) I'm sure you didn't enjoy my first piano recital. To have would have
been ex/raon/inary.6
The crucial constraint on accent placement does not involve VP deletion:
the relevant factors are (a) phonological repetition, (b) identity of sense,
(c) sentence-final position.
As additional support for their argument that VP deletion is involved in
the constraint they identify, Z. and L. claim that 'verbs other than have and
be occur freely in their stressed infinitival forms before gaps — gaps created
by rules other than VP deletion', as in (13) and (14) (their [24] and [25]):
(13) Everyone thinks that Millie will pass the exam, but I don't even
think she'll try.
(14) I asked Norman why he was sobbing but he wouldn't say.
Following the argument used for (4) to (8), we see that such verbs do not
occur FREELY with the accent, since the occurrence of the accent is subject
to constraints (a) and (b):
(13') ?Everyone thinks that Millie will (try) to pass the exam, but I don't
think she'll try.
(14') ?I asked Norman to (say) why he was sobbing but he wouldn't say.
A similar argument applies to Z. and L.'s treatment of do. They claim
that while nonfinite do cannot bear the accent in a VP remainder, as in
(15) (their [31]), it can bear the accent before gaps created by other
movement rules, as in (16) (their [32]):
(15) Trace theorists will solve these problems. Well, at least they
could do.
*could do.
(16) We must do some of these exercises. Which shall we do!
Though Z. and L. might argue that our three constraints above fail to
apply to the remaindered VP do, this would be incorrect, since (16) does
not in fact contain a VP remainder. The second instance of do is not a
remainder of do some of the exercises, but a proform do, substituted for
the idiomatic do of the first part of the sentence, as is shown by comparing
(16) to (17) and (18):
(17) We must do some exercises.
mend some shoes.
bake some cakes. How many shall we do?
write some letters.
cut down some trees.
We have shown that our constraints are quite general in their applica-
tion. It is therefore desirable to use the same constraints to account for the
unacceptability of the accent on to, have, be, been, and being when they
are remainders of VP deletion. To do so, we propose that the remainder
of VP deletion functions phonologically as a proform for the VP. Thus in
(2"), to have functions as a proform for the VP enjoy my first piano recital:
(2") I'm sure you didn't enjoy my first piano recital. You'd have to be
crazy to have.
In this respect, the VP remainder is directly comparable to pronouns such
as // in (2'") and other proforms where givenness precludes accent
placement:
(2"') I'm sure you didn't enjoy my first piano recital. You'd have to be
crazy to have enjoyed it.
Just as it is ungrammatical to locate the accent on it in (2"') without
destroying the coreference relation between anaphor and antecedent (my
first piano recital), so it is ungrammatical to locate the accent on the
rightmost element of the VP remainder in (2"), because this would signal
the VP remainder as new information, thus destroying the anaphoric
relationship between VP remainder (to have) and its antecedent enjoy my
first piano recital. The role of accent placement in destroying conventional
coreference interpretations is illustrated by comparing (20) to (20'):
(20) John and Mary were waiting in the room when Frank came in.
John kissed Frank and then Mary kissed him. [him = Frank]
(20') John and Mary were waiting in the room when Frank came in.
John kissed Frank and then Mary kissed him. [him = John ]
In (20') the location of the accent on the proform signals something like
look for a referent other than the conventional one for this proform. For
this reason, we hypothesize that the accent cannot occur on the rightmost
element of a VP remainder, as this would signal something else: i.e. don't
identify this anaphor with the anetecedent conventionally assigned to it; find
another antecedent VP.1 This is incoherent in all the sentences cited by Z.
and L., since (a) there is no other suitable antecedent and (b) the
conventional interpretation is what the speaker is aiming for.8
But the accent must go somewhere. This requirement is responsible for
the apparent failure of the proform deaccenting hypothesis in certain
cases we shall now consider. In the light of these data we can conclude
that it is ONLY the rightmost element of the VP remainder that functions
phonologically as a proform: so the accent can be shifted leftward within
the VP remainder without destroying the desired (conventional) reading.
This accounts for the facts that Z. and L. found so puzzling, illustrated in
(21) and (22):
(21) I didn't see the exhibit last time, but this time I'm likely to see it.
likely to see it.
*likely to see it.
likely to.
likely to do so.
*likely to do so.
(22) I'm sure you didn't enjoy my first piano recital. You would have
had to be extraordinary to have enjoyed it.
to have enjoyed it.
*to have enjoyed it.
to have.
*to have.
to.
*to.
to have done so.
*to have done so.
We have now accounted for all Z. and L.'s data with infinitive have, be, to,
been, and being. The problem is that our explanation appears to be too
powerful, predicting that some of the grammatical sentences produced by
Z. and L. will be ungrammatical: specifically, sentences with finite
auxiliaries that precede a gap created by VP deletion.
the tag, because accent on the pronoun would destroy the desired
coreference relationship:
(27) John'Sj done it, has hej?
(28) *John'Si done it, has he{!
(29) John'Si done it. Has he?.
There remain many instances of finite auxiliaries bearing the accent before
a gap created by VP deletion. In such cases, the finite auxiliary can bear
the accent just because it is the bearer of the new information content of
that part of the utterance. On its own, the auxiliary may convey new
information of one or more of the following kinds: (i) affirmative vs.
negative, (ii) time reference, (iii) modality. If the new information of the
utterance consists of one (or more) of these three, it will frequently be the
case that the leftward shift of the accent is blocked, because there is no
other element onto which the accent, and thus the information focus, can
be shifted without depriving the auxiliary of its 'new information' status.
Take the case of negation. In the case of NONFINITE forms, we saw earlier
that the accent can be shifted leftward, thus marking the VP remainder as
given, while at the same time retaining the new information that the
negation of the preceding sentence is being reversed:
(30) I'm sure you didn't enjoy my first piano recital. You would have
had to be crazy *to have.
to have.
The situation is perhaps clearer when the new element is the negative:
(31) I'm sure you enjoyed my piano recital. You would have had to be
crazy *not to have.
not to have.
In (31), leftward shift takes the accent off the given VP remainder, and at
the same time marks the negative particle not as new. In (32), on the other
hand, the accent cannot be shifted off the auxiliary if the reversal of
negation is to be retained as the new information.
(32) I don't think the machines work, but Ann thinks they do.
In fact, the case of (32) is complicated by Z. and L. omitting to mark the
accent in the first half of the sentence; we must assume a 'neutral' accent
placement on the last lexical item. (32') to (32'") represent the possible
leftward shifts:
(32') *I don't think the machines work, but Ann thinks they do.
(32") *I don't think the machines work, but Ann thinks they do.
(32'") *I don't think the machines work, but Ann thinks they do.
(32') and (32") are unacceptable because they violate the lexical repetition
constraint stated above. (32'") is pragmatically incoherent, since marking
Ann as new by accent placement implies that the first part of the sentence
is about the person holding the opposite view to Ann, whereas the neutral
accent in the first part carries no such implications. The accent must
therefore be on do if the reversal of negation is to be presented as the new
information content of the second part of the sentence. Of course, the
accent can be elsewhere in the first part, altering the situation radically, as
in (32a) and (32b):
(32a) / don't think the machines work, but Ann thinks they do.
(32b) / don't think the machines work, but Ann thinks they do.
The location of the second accent in (32a) is coherent, since the first
accent indicates that the first part of the sentence is about who is holding
an opinion. (32b) is a more interesting case. Here, once again, the second
accent on do implies that the second part is about the working vs.
nonworking of the machines and that Ann is contextually given, while the
first part is about who holds an opinion. Having looked at some of the
many possible realizations of (32), we can sec that, given neutral accent
placement in the first part of the sentence, the speaker has no alternative
but to place the accent on the finite auxiliary if he is to present reversal of
negation as the new information. Shifting the accent leftward would
either violate the lexical repetition constraint (which, as we saw, is
stronger for nouns than for verbs) or else destroy the pragmatic coherence
of the utterance by presenting as new information an item (Ann) which
does not fulfil the hearer's expectations, set up by the location of the first
accent, of what the second part will be about.
Conclusions
they are mediated by lexical and /or syntactic systems). We have shown
that such an approach provides a more adequate description of accentua-
tion than is possible within the narrow generative model espoused by
Zwicky and Levin.
Notes
1. A version of this paper was given to the York University Post-graduate Linguistics
Seminar in 1981. Thanks to those who made useful comments; in particular, thanks to
Steve Harlow, specifically for the analysis of do and generally for help with semantic
terminology; to John Kelly for phonological guidance; and to anonymous reviewers.
2. Z. and L. state that they are dealing with 'contrastive stress' on 'contrastive VPs'
without explaining what they mean by these extremely problematic terms (see Chafe
1976; Ladd 1980; and Wells forthcoming, among others). Happily, the notion of
contrastivity is of no significance for their argument.
3. The unacceptability of (3') is dependent on the accent being located on 'chosen' in the
FIRST sentence. If the accent is on 'not', the sentence becomes acceptable: º wouldn't be
upset not to be chosen. But on the other hand I wouldn't be upset to be chosen.' This
conforms to the general principle proposed later: 'chosen' is presented as given,
cataphorically, in the first sentence, and so can be presented as new in the second part
and bear the accent. Z. and L.'s data is often difficult to interpret, as they omit to mark
the location of the accent in their context sentences. We remedy this deficiency in our
own examples and supply a 'neutral' for Z. and L.'s. Accents are marked by italicizing
the relevant syllable.
4. Unacceptability here depends not on phonological identity but on identity of sense.
Consider (i) and (ii) in which synonyms are substituted for the repeated items in (2')
and (3'), and which are also unacceptable:
(i) ?I'm sure you didn't enjoy my first piano recital. You'd have to be crazy to have
liked it.
(ii) ?I wouldn't be upset not to be chosen. But on the other hand I wouldn't be upset to
be se/ected.
5. It would be interesting to see statistics on the frequency of though inversion sentences in
SPOKEN English. They seem to us to be quite rare; our intuitions about the acceptability
of different accent placements are consequently uncertain.
6. We have observed a few cases parallel to 10 through 12 in British English, though it is
possible that there are dialect differences here. The following example is taken from a
TV documentary:
(iii) It's come just in time for Sarah, who with her blisters admits that she'd have
stopped if only there'd been somewhere to — but amongst peat and driving rain
she'd had to go on.
Invoking such examples merely underlines the difficulty of using intuitive data,
constructions that parallel the pronominal anaphora of (20'), and so there are no test
cases for the parallelism between the two types of anaphora, with regard to the breaking
of conventional coreference assignment. Our argument must therefore rest on the
phonological deaccenting that occurs both on conventionally referring pronominals and
on conventionally referring VP remainders.
References