paper on relative boundary strength

Here’s the revised version of our paper on gradient boundary strength and disambiguation, soon to be presented at speech prosody:

Wagner, Michael & Serena Crivellaro: Relative Prosodic Boundary Strength and Prior Bias in Disambiguation

Abstract: Previous research found that the relative rather than the absolute size of prosodic boundaries is crucial in disambiguating attachment ambiguities [1, 2]. Furthermore, relative categorical differences matter whereas merely quantitative ones do not [1]. This paper presents further evidence that relative boundary strength is indeed what is crucial, but, contrary to earlier findings, gradient quantitative differences in boundary rank affect parsing decisions in gradient ways. Furthermore, varying the plausibility of a given reading in a given context shifts the perceptual boundaries between different phrasings such that quantitatively stronger prosodic cues are necessary to counter-act a prior bias against it.

on linguistic interfaces II

The call for papers for the second conference on linguistic interfaces was posted. The deadline is April 30th. Here’s part of the conference description:

A full description of our knowledge of language must include reference to several different components, each with its own particular properties. These components must interact with each other, and with a lexicon, which we may think of as a system of stored associations between pieces of information pertaining to many of the above components. In recent years, the study of the interaction between these different levels of linguistic knowledge has attracted increasing interest. The nature and extent of the interaction of different linguistic modules is a central question to be addressed by a modern theory of linguistic knowledge. […]

heads-up: statistics workshop

Statistics Workshop on Logit Mixed Models, May 3-4 2010 at McGill

The gripp reading group and the CRLMB are organzing a statistics workshop on logit mixed models. Florian Jaeger (University of Rochester) and his lab will give several tutorials on this and related topics on May 3 and 4 (we might have a 3rd day though). The workshop is co-sponsored by the CRLMB, prosody.lab, the Mcgill Infant Development Cluster, and the PoP lab, and is organized by Aparna Nadig, Kris Onishi and me.

Further information will be posted in early March. Hope to see you there!

if worst comes to worst

I was about to write if worse comes to worst in an email, and then realized that I’m actually not sure what the correct idiom exactly is. The difference between worse and worst is hard to hear, and it’s not so clear that it has a transparent compositional meaning, so neither sound nor meaning really help. A quick Google vote seemed to resolve this, but not in an entirely crisp way:

“worst comes to worst” – 149.000 hits
“worse comes to worst” – 47.800 hits
“worst comes to worse” – 578.000 hits
“worse comes to worse” – 7.370.00 hits

So I went with ‘worse comes to worse.’ However, the Merriam Webster’s Dictionary of English usage states that the original phrase is ‘if the worst comes to the worst,’ first used in 1597, and then goes to say that

More …

did you say toyota or toyoda?

The name of the car company and that of its president and CEO (a grandson of its founder Kiichiro Toyoda) are homophonous (or at least nearly homophonous)–in North American English, that is. Both [t] and [d] become flaps intervocalically after a stressed syllable.

It’s interesting though to see what people do when they want to emphasize which one of the two words were intended. According to various people I’ve polled (some of them currently taking my experimental class at the moment), one way to disambiguate is to shift stress to the last syllable and thus bleed the application of flapping.

This is an example of focus within-word focus. In his 1961 paper on contrastive accent and contrastive stress Dwight Bolinger describes another instance:

In a New Yorker cartoon a man stands upside down, with feet on the ceiling, in a psychiatrist’s office. The psychiatrist says to the man’s wife, ‘In a case of this kind, Mrs. Hall, our first concern is to persuade the patient that he is a stalagmite’ (last syllable underlined in original).

There’s a nice paper byRon Artstein that discusses the semantics of this kind of focus below he word level.

It’s conceivable though that a speaker might shift stress even in the absence of such a contrasting minimal pair, in order to clarify the spelling (or underlying form?) of a word. If this way of avoiding flapping became a regular process, it would constitute a violation of a generalization that is otherwise very robust, (the ‘two-many-solutions-problem’, see Donca Steriade’s paper on the P-Map and
Lev Blumenfeld’s 2006 thesis): Prosodically conditioned lenition-processes such as foot-internal flapping are not mirrored by hypothetical processes that would shift stress for the purpose of avoiding the lenition to apply. This is surprising under an optimality-theoretic analysis of such processes that uses the interaction of markedness and faithfulness constraints, since shifting stress would be another way to avoid the markedness-violation incurred by intervocalic post-tonic [t] or [d], so it would be easy to create a correct ranking for such a grammar. However, languages don’t ever seem to pick this solution. They either stay faithful or lenite.

So does the English strategy to emphasize the underlying (or at least orthographic) contrast between ‘Toyota’ and ‘Toyoda’ show that this generalization is not really true? Maybe not: for all we know, the stress shift observed here might be an acrobatic act that savvy users of English use to disambiguate otherwise homophonous words based on their own phonological knowledge, but not something that could ever become a regular grammatical process that could be generalized to all flapping environments.

On the other hand, maybe such a process is simply unlikely to be develop: Languages might conserve a contrast (as in languages that don’t have flapping, say British English) or lenite–but under what scenario would a language develop a stress shift that serves the purpose of avoiding a neutralization brought about by a lenition process which, however, could never be observed in the first place if there were a stress shift? Developing such a pre-emptive phonological process might simply be an unlikely scenario for a language change.

Shifting stress is different from other forms of enhancement (which one could argue are similarly pre-emptive) since it comes at the cost of changing a property that itself is contrastive in English, word-stress. Another difference might be that shifting stress does not exaggerate a tendency that is already present with the original contrast (That’s an empirical question: Could a hyper-articulated rendition of [toi!] in French at times be aspirated? Could a hyper-articulated British English rendition of [water] at times shift stress to the last syllable? It seems to me the former is much more likely than the latter, but maybe I’m wrong).

So it’s not so obvious that there is a ‘two-many-solutions’ problem in this case. Of course, this type of explanation might only be feasible in a small subset of cases in which the ‘two-many-solutions’ problem has been observed.

[some edits on February 9, 2010]

tie in stockholm, deadline feb 15

The TIE conference series brings together researchers working on different aspects of tone and intonation in a wide variety of languages. Abstracts for papers and posters are welcome on any topic within these areas.

conference homepage

wccfl

The WCCFL program has been posted. There’s lots of interesting papers on prosody on the program, including some from McGill: by David-Etienne Bouchard (on clitic in French), Öner Öçelik and Miho Nagai (Syntax-Prosody in Turkish), Hyekyung Hwang, Moti Lieberman, Heather Goad, & Lydia White (on prosodic diambiguation), Walter Pederson (on ‘again’), and yet another one by Heather Goad and Lydia White (on prosodic transfer). There’s also a talk by Nikola Predolac, who is just finishing his thesis at Cornell, on givenness and the scope of focus operators in Serbian. Lots of other interesting prosody stuff on the program, including an invited talk by Duane Watson. Can’t believe I can’t go…

speech prosody

Speech Prosody will take place in Chicago this year. There’s a special session on Experimental Approaches to Focus, organized by Yi Xu which promises to be interesting.

prosodylab will be represented at speech prosody with two posters in the main session. drop me an email if you’re interested in the drafts:

M. Wagner, M. Breen, E. Flemming, S. Shattuck-Hufnagel, & E. Gibson: Prosodic Effects of Discourse Salience and Association with Focus.

Abtract: Three factors that have been argued to influence the prosody of
an utterance are (i) which constituents encode discourse-salient
information; (ii) which constituents are contrastive in that they
evoke alternatives; and (iii) which constituents interact with
the meaning of focus operators such as only (i.e., they ‘associate’
with focus). One challenge for a better understanding of
these factors has been the difficulty of finding a way to evaluate
hypotheses quantitatively, since individual variation in productions
is often large enough to wash out experimental effects.
In this paper, we apply a methodology introduced in [1] which
regresses out subject and item variation, uncovering otherwise
hidden prosodic patterns that illustrate how the three factors interact
in sentences containing single or multiple foci.

Wagner, Michael & Serena Crivellaro: Relative Prosodic Boundary Strength and Prior Bias in Disambiguation

Abstract: Previous research found that the relative rather than the absolute
size of prosodic boundaries is crucial in disambiguating
attachment ambiguities [1, 2]. Furthermore, relative categorical
differences matter whereas merely quantitative ones do not
[1]. This paper presents further evidence that relative boundary
strength is indeed what is crucial, but, contrary to earlier findings,
gradient quantitative differences in boundary rank affect
parsing decisions in gradient ways. Furthermore, varying the
plausibility of a given reading in a given context shifts the perceptual
boundaries between different phrasings such that quantitatively
stronger prosodic cues are necessary to counter-act a
prior bias against it.