HOWL-4Abstracts for talks and Tutorials |
 |
Tutorial Abstracts
FRIDAY, 10/12/07 – TUTORIALS (SALON C)
Tutorial: Structure and Statistics in the Input
Colin Wilson, UCLA
The HOWL 4 invited phonology talks address a wide range of issues, with two major unifying threads: the existence and acquisition of statistical generalizations over phonetic, phonological, and lexical structures; and the ways in which separate generalizations interact, as measured in natural performance or through various experimental tasks. This tutorial reviews the main types of linguistic structure that are relevant for phonetics and phonology (e.g., gestures, cues, segments, syllables) and existing methods for assessing and learning statistical phonological generalizations and their interaction (e.g., positional frequency tracking, interactive activation, maximum entropy modeling, and stochastic ranking). The emphasis will be on understanding both the high-level goals and assumptions of each method and the low-level mechanics of applying each one to phonological data. (The much more difficult problem of constructing conceptual or empirical arguments that favor one type of model over another will be left to the invited speakers.) There are no particular prerequisites or reading requirements, but the following is a short list of introductory-level articles that are broadly relevant to the tutorial and the phonology session:
Jusczyk, P.W. (2007) Speech Perception, Development of. Encyclopedia of Cognitive Science. John Wiley & Sons.
Nowak, M.A., N.L. Komarova, and P. Niyogi (2002) Computational and evolutionary aspects of language. Nature, 417, 611-617.
Pierrehumbert, J. (2001) Stochastic phonology. GLOT, 5, 1-13.
Saffran, J. (2003) Statistical Language Learning: Mechanisms and Constraints. Current Directions in Psychological Science, 12, 110-114.
Tutorial: Autonomy versus interaction
John Kingston
University of Massachusetts
In this tutorial, I will review the principal evidence presented by proponents of autonomous and interactive models of speech perception, as background and context for my presentation. This review is based on a selection of arguments from McClelland & Elman (1986), a much closer look at the evidence presented in Elman & McClelland (1988), Pitt & McQueen (1998), and Magnuson, McMurray, Tanenhaus, & Aslin (2003), and then once again a more selective evaluation of arguments autonomy in Norris, McQueen, & Cutler (2000) and for interaction in McClelland, Mirman, & Holt (2006). These papers focus on two questions: when does knowledge of what is a word influence the evaluation of the signal and are effects that look like the influence of lexical knowledge actually the influence of knowledge about how likely one segment is to follow another (AKA transitional probabilities)? The tutorial will wrap up by examining evidence which shows that knowledge of phonological constraints influences speech perception independently of lexical or statistical knowledge (Moreton, 2002; Coetzee, 2005).
References
Coetzee, A. W. (2005). The obligatory contour principle in the perception of English, S. Frota, M. Vigario, & M.-J. Freitas (eds.) Prosodies: With Special Reference to Iberian Languages, (pp. 223-245), Berlin: Mouton de Gruyter.
Elman, J. L., & McClelland, J. L. (1988). Cognitive penetration of the mechanisms of perception: Compensation for coarticulation of lexically restored phonemes, J. Mem. Lg., 27, 143‑165.
Magnuson, J. S., McMurray, B., Tanenhaus, M .K. & Aslin, R. N. (2003). Lexical effects on compensation for coarticulation: the ghost of Christmash past, Cognitive Science, 27, 285-298.
McClelland J. L., Elman J. L. (1986). The TRACE model of speech perception, Cognitive Psychology, 18, 1-86.
McClelland, J. L., Mirman, D., & Holt, L. L. (2006). Are there interactive processes in speech perception? Trends in Cognitive Science, 10, 363‑369.
Moreton, E. (2002). Structural constraints in the perception of English stop-sonorant clusters, Cognition, 84, 55-71.
Norris, D., McQueen, J. M., & Cutler, A., (2000). Merging information in speech recognition: Feedback is never necessary, Behav. Brain Sci., 23, 299-370.
Pitt, M. A., & McQueen, J. M. (1998). Is compensation for coarticulation mediated by the lexicon? J. Mem. Lg., 39, 347-370.
Tutorial: Core Systems of Number
Lisa Feigenson, JHU
What representations underlie our ability to think and reason about number? Are there qualitative differences in our representations of quantities such as “exactly one,” “approximately ten,” and “two hundred and thirty six?” In this tutorial I will review evidence that some numerical abilities are widespread and can be observed in adults, infants, and other animal species. Recent behavioral, neuroimaging, and neuropsychological evidence suggests that these ontogenetically and phylogenetically shared abilities rest on two core systems for representing number. Performance signatures common across development and across species implicate one system for representing large, approximate numerical magnitudes, and a second system for precisely representing small numbers of individuals. These systems account for our basic numerical intuitions, and serve as the foundation for the more sophisticated numerical concepts that are a uniquely human creation. Yet neither of these has the representational precision of exact integer concepts, which may only be attained by a subset of human adults and which have been suggested to require language for their construction.
Tutorial: Quantifiers, Plurals and the Semantics of Number
Graham Katz, Georgetown University
How do we talk about many things at the same time? Common ways include the use of quantificational expressions such as "every boy", "most papers" and "few jobs", the use of plural nominals, such as "the cats" and "large whales" and "they", and the use of conjoined and group-referring expressions such as "Mary and Sue" or "the committee". In this tutorial we will survey standard formal analyses of these expressions that relate, in one way or another, to more than one object, discussing at an introductory level the ontological commitments that modern theories of plurality (Link 1983), group reference (Winter 2002) and kind reference (Krifka et. al. 1995) are driven to, and exploring some of the basic issues associated with the heavily studied domain of generalized quantification (Keenan & Westerstahl 1997). We will also consider the semantic import of grammatical number (Sauerland 2003).
Chierchia, G. (1998). Plurality of mass nouns and the notion of “semantic parameter”. In S. Rothstein, (ed.), Events and Grammar, pp. 53–103. Kluwer, Dordrecht, Netherlands.
Krifka, M., F. J. Pelletier, G.N. Carlson, A. ter Meulen, G. Chierchia & G. Link (1995). Genericity: An Introduction. In G.N. Carlson and F.J. Pelletier, The Generic Book. Chicago: University of Chicago Press. pp. 1-124.
Sauerland, U. (2003). A new semantics of number, in Proceedings of SALT 13, CLC Publications (Cornell University), Ithaca.
Link, G. (1983). The logical analysis of plurals and mass terms: A lattice-theoretical approach. In Rainer Bauerle, Christoph Schwarze, and Arnim von Stechow (eds.), Meaning, Use and Interpretation of Language, pp. 302-323. Berlin: de Gruyter.
Winter (2002). Atoms and Sets: a characterization of semantic number. Linguistic Inquiry 33:493-505.
Keenan, E. and Westerstahl, D. (1997). Generalized quantifiers in linguistics and logic, in J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language, Elsevier, Amsterdam, 837-893.
Talk Abstracts
SATURDAY, 10/13/07 – SYNTAX/SEMANTICS OF NUMBER (SALON C)
On the conceptual role of number marking in French Based Creoles
Viviane Deprez
Rutgers University
The syntactic position of number marking (#)—on N or D—has been argued to determine three aspects of nominal syntax: adjective-order, noun-omission and the licensing of Bare Argument Nominals (BAN) (Bouchard 2002). This paper shows that for a whole class of languages, i.e. the French bases creoles number position is insufficient to make correct predictions for these nominal properties and argues that better results can obtain if the conceptual role of number marking is also taken into account.
Number interacts with two distinct processes in language: Individuation and Quantization. Only the former will be of concern in this talk. In Individuation, an operation that maps a nominal concept to the individual objects that materialize it, the role of number is to provide a criterion of division mapping given properties onto distinct instances. But Individuation, the talk argues, can also be achieved independently of number: contextual localization provides an alternative criterion of division. That is, I similarity to how objects have been argued to be individuated through their properties or their location in space (Xu and Carey 1996), so, I propose, can a kind be mapped onto its realizations through number division/atomization or spatial location. Languages opt for different modes of individuation and this choice has empirical consequences on nominal syntax. Number position, I show, matters only for languages with Number Individuation. Accordingly, not all languages that mark number primarily on D have equivalent properties and some are correctly predicted to freely allow BAN.
The Importance of Being Arbitrary: Causes of Variation in Numerical Cognition and Representation
Peter Gordon, Columbia University
In this presentation, I explore the relationships between cultural and individual variation in numerical representation. I explore the causes and consequences of developing systems of individuation and quantification that contact either exact numerical representation versus those that rest on their laurels with approximate formats. I will consider the view of approximate representations of number as a developmental precursor of exact numerical representation and I consider two hypothetical mechanisms by which this change occurs: 1. The prototype-plus-trigger hypothesis, which claims that approximate representations are numerical concepts that require certain triggering experiences (e.g., counting) to make them exact, and 2. The metamorphosis hypothesis, which posits a more radical change in representational format from approximate to exact enumeration. I propose that one key role for language in affecting metamorphic representational change is its arbitrary nature, which forces the development of arithmetic theory-like structures to develop in support of the meaningful representation of the domain of number. I suggest that non-arbitrary iconic systems of counting are defective and insufficient to develop true exact numerical systems of representation.
Beyond Truth Conditions: Towards a Psycho-semantics for most
Jeff Lidz, University of Maryland & Justin Halberda, JHU
How is the meaning of the word "most" related to human capacities for detecting and comparing numerosities? One might think the answer is both obvious and given by our best semantic theories: we understand most in terms of a capacity to compare cardinal numbers. We provide some initial evidence against this view on the basis of experimental evidence from children's understanding and from psychophysical experiments of adults' understanding. First, we show that even children who have no exact numerical competence for large numbers are able to accurately compute the meaning of most.
Second, we show that under conditions that make it impossible for adults to access exact cardinality, they are still highly accurate in judging the truth of a sentence containing most. Finally, we argue against a view that treats "most" as derived from a simple vote-counting algorithm that yields cardinality relations without making reference to cardinality (via Hume's Principle) and in favor of a view that involves the approximate number system (analog magnitudes) found in humans and other animals.
Language and Precise Number: The Neurobiology of Quantification
Robin Clark
University of Pennsylvania
We examine the understanding of sentences containing a single quantifier.
Subjects were asked to make truth value judgments of a sentence relative to
a model (the latter presented visually), where each sentence contained a
single quantifer which could be an Aristotelean (e.g., "all", "some",
"no"), a cardinal determiner ("at least n", where n > 0), a parity
determiner (e.g., "an even/odd number of") or a majority determiner (e.g.,
"more than half"). The judgment marshalled parts of the parietal lobe
normally associated with number processing; furthermore, a neuroanatomical
difference was noted between first order and higher order quantifiers.
Based on these observations, we compared different populations ---corticobasal degeneration (CBD) patients, fronto-temporal dementia (FTD) patients and
Alzheimers (AZ) patients---with respect to their behavior on various types of quantifiers. The results suggest that understanding of sentences containing quantifiers crucially involves number sense.
SUNDAY, 10/14/07 – HOW THE COGNITIVE SYSTEM SHAPES PHONOLOGY (SALON C)
John Kingston
University of Massachusetts
Does what you know about your language influence how you perceive speech? The answer to this question is undoubtedly, “yes,” if for no other reason than that we often fail to distinguish sounds from another language that do not contrast in our own. A better question is therefore, when, in the course of processing an incoming speech signal, does what you know come to influence how you perceive the acoustic contents of that signal? Two competing answers may be found in the literature: proponents of interactive models argue that linguistic knowledge influences perception immediately (McClelland & Elman, 1986; Elman & McClelland, 1988; Dupoux, Kakehi, Hirose, Pallier, & Mehler, 1999; Dehaene-Lambertz, Dupoux, & Gout, 2000; Dupoux, Pallier, Kakehi, & Mehler, J., 2001 McClelland, Mirman, & Holt, 2006), while proponents of autonomous models argue instead that the initial auditory evaluation is uninfluenced by linguistic knowledge (Norris, McQueen, & Cutler, 2000; Kingston, 2005).
I will present the results of two sets of experiments which test the prediction of interactive models that linguistic knowledge, specifically knowledge of what is a word, improves the evaluation of the signal’s acoustic contents and thus makes stimuli from a word-nonword continuum more discriminable. Both sets of experiments disconfirm this prediction. The first set of experiments also shows that lexical biases are unaffected by the speed of the listener’s response, again disconfirming the prediction of interactive models, as implemented in TRACE, that lexical biases would grow with response time. The second set shows that the response bias induced by the target segment’s neighbors is independent of that induced by lexical knowledge, a result which is compatible with the autonomist assumption that there is a level of processing separate from one when linguistic knowledge is applied. These results are pieces of a larger argument for an autonomous level of auditory processing that transforms the acoustic contents of the signal in ways that perceptually unite variable pronunciations of distinctive feature values and divide the signal into segment-sized bundles of properties.
References
Dehaene-Lambertz, G., Dupoux, E., & Gout, A. (2000). Electrophysiological correlates of phonological processing: A crosslinguistic study, J. Cognitive Neuroscience, 12, 635-647.
Dupoux, E., Kakehi, K., Hirose, Y., Pallier, C., & Mehler, J. (1999). Epenthetic vowels in Japanese: A perceptual illusion, J. Exptl. Psych.: Human Perception and Performance, 25, 1568-1578.
Dupoux, E., Pallier, C., Kakehi, K., & Mehler, J. (2001). New evidence for prelexical phonological processing in word recognition, Language and Cognitive Processes, 16, 491-505.
Elman, J. L., & McClelland, J. L. (1988). Cognitive penetration of the mechanisms of perception: Compensation for coarticulation of lexically restored phonemes, J. Mem. Lg., 27, 143‑165.
Kingston, J. (2005). Ears to categories: New arguments for autonomy, S. Frota, M. Vigario, & M. J. Freitas, (eds.) Prosodies: With Special Reference to Iberian Languages, (pp. 177-222), Berlin: Mouton de Gruyter.
McClelland J. L., Elman J. L. (1986). The TRACE model of speech perception, Cognitive Psychology, 18, 1-86.
McClelland, J. L., Mirman, D., & Holt, L. L. (2006). Are there interactive processes in speech perception? Trends in Cognitive Science, 10, 363‑369.
Norris, D., McQueen, J. M., & Cutler, A., (2000). Merging information in speech recognition: Feedback is never necessary, Behav. Brain Sci., 23, 299-370.
Graded Constraints in English Word Forms
Jay McClelland (Stanford) and Brent vander Wyk (CMU)
We will describe a graded constraint theory of English word forms that addresses the distribution of forms in the lexicon and the goodness judgments given by native speakers. The theory is applied to the rhymes of English monosyllabic mono-morphemes (items like 'cat', 'hold' and 'clamp'). Within a template specifying possible rhymes, a number of graded constraints are identified. For example, in rhymes containing at least one stop consonant, there is a graded constraint favoring short vowels, a graded constraint favoring unvoiced over voiced obstruents, a constraint favoring coronal articulation, and a constraint against added embellishments such as a nasal, fricative, liquid, or second stop consonant (as in 'apt').
Each constraint affects the goodness of a rhyme type in a graded, cumulative fashion. Occurrence rates of different types of rhymes in the language conform closely to the predictions of both non-parametric and parametric versions of the theory. By adding a cut-off threshold, the theory can explain with good accuracy which types of rhymes occur at all and which do not occur, although both linear and interaction terms are necessary to give a complete account. The theory also accounts well for native speaker's judgments of the relative goodness of different rhyme types, although there are subtle differences between the patterns of occurrence and the patterns of judgments. The effects of constraint violations on the durations of spoken word forms indicate that each constraint violation increases the duration of a spoken word for in essentially a linear, additive fashion.
Realistic Input and the Processing of Phonetic Detail: Moving Beyond CV Syllables
Lisa Davidson
New York University
Research in cross-language speech perception has led to several influential theories about how the phoneme inventory and phonological grammar of one’s native language affect the processing of non-native input (e.g. Flege 1995, Best 1995, Kuhl 1993). However, the studies informing these theories often rely on results from stimuli that are monosyllabic, have simple syllable structure, or are devoid of other linguistic information, such as meaning (e.g. Werker & Tees 1984, Polka 1991, Best, McRoberts & Goodell 2001). Results based on simplistic stimuli may not adequately represent how cognitive factors such as the phonological grammar or lexical representations affect the processing of non-native speech in more realistic situations such as loanword adaptation or second language acquisition.
In this talk, I will present results from two studies designed to better understand the cognitive demands present in more realistic language-learning situations. The first experiment examines the role of meaning in learning phonetically similar but phonologically distinct words. Participants’ ability to learn a contrast between CC-initial and CəC-initial words is examined with a picture-word learning task. In the first part, American English listeners were trained on word-picture pairings of words containing a phonological contrast between CC and CəC sequences, but which were not minimal pairs (e.g., [ftake], [fətalu]). The second part of the study consisted of word-picture pairings containing minimal pairs (e.g., [ftake], [fətake]) to investigate whether listeners are more likely to attend to the presence of the schwa when it alone carries a difference in meaning. Results suggest that listeners accept a great deal of variability in the signal unless they are forced to attend to phonetic detail for lexical discrimination.
The second study examines how grammatical factors and cognitive resources such as language background, word length, and order of stimulus presentation affect cross-language speech perception. Catalan, American English, and Russian speakers were presented with two types of word pairs: long, CCVCV~CəCVCV (e.g. /dbaka~dəbaka/) and short, CCV~CəCV (/dba~dəba/) in an AX discrimination task. Results based on the length dimension indicate that increasingly realistic multisyllabic stimuli tax a listener’s resources, presentation order suggests that the stability of the representation of a new word (even if only temporary) depends on the phonotactic legality of the word, and native language background underscores how a speaker’s phonological grammar contributes to the perception of what is otherwise the same stimuli for all participants.
What is the Relation of Cross-Linguistic Patterns and Language Development?
LouAnn Gerken
University of Arizona
The foundation for much of the study of linguistics and language development is a pair of observations: similarities exist in the patterns observed among human languages, and children’s immature linguistic forms often bear some resemblance to typical (unmarked) patterns observed cross-linguistically. There are at least two accounts for these observations. The traditional account is that both cross-linguistic patterns and children’s immature forms reflect a single factor – the human biological endowment for language. However, another account is also possible. On this view, similarities among human languages reflect a set of interacting factors, including constraints on what humans are able to conceive, learn, remember, perceive, and produce. This multi-factor account provides an explanation for the observation that forms that are unmarked cross-linguistically are often the most frequent forms within a particular language, which in turn influence language development.
My talk presents two lines of research. One line asks whether children’s immature word forms more closely resemble unmarked forms or the forms that are statistically reliable in their own input. The answer appears to be that input is more influential, admitting the possibility that the resemblance observed between cross-linguistic markedness and children’s immature forms is due to learning. The second line of research asks whether infants are able to learn a broader range of generalizations than observed in cross-linguistic data. The answer appears to be that they are, but that over development, the generalizations that they are able to learn come rapidly to be constrained by statistically reliable patterns in their input, most of which are also found cross-linguistically. The talk ends with a discussion of how the fields of linguistics, language development, computational modeling, and machine learning might interact to further explore the relation between cross-linguistic patterns and language development that is suggested by the developmental research.