colloquia

Phonological Models in Automatic Speech Recognition

Karen Livescu
Toyota Technological Institute at Chicago
Location: Cobb 201
Time: 3:30pm

Abstract:

The performance of automatic speech recognizers varies widely across contexts. Very good performance can be achieved on single-speaker, large-vocabulary dictation in a clean acoustic environment, as well as on small-vocabulary tasks with fewer constraints on the speakers and acoustics. One domain that is still elusive is that of spontaneous conversational speech. This type of speech poses a number of challenges, among them extreme variation in pronunciation. I will describe efforts in the speech recognition community to characterize and model pronunciation variation.

The most thoroughly studied approach is augmentation of a phonetic pronunciation lexicon with phonological rules. Despite successes in a few domains, it has been surprisingly difficult to obtain significant recognition improvements by including those phonetic pronunciations that appear to exist in the data. I will advocate an alternative view: that the phone unit may not be the most appropriate for modeling the lexicon. I will describe approaches using both larger (e.g. syllable-sized) and “smaller” (e.g. articulatory) units. In the class of “smaller” unit models, ideas from articulatory and autosegmental phonology motivate multi-tier models in which tiers have semi-independent behavior. I will present a particular model in which articulatory features are represented as variables in a dynamic Bayesian network.

Non-phonetic pronunciation models can involve significantly different model structures than those typically used in speech recognition, and as a result they may also entail modifications to other components such as the observation model and training algorithms. At this point it is not clear what the “winning” approach will be. The success of a given approach may depend on the domain or on the amount and type of training data available. I will describe some of the current challenges and ongoing work, with a particular focus on the role of phonological theories in statistical models of pronunciation (and vice versa?).

Getting ‘Better’: On Comparative Suppletion and Related Topics

Jonathan Bobaljik
University of Connecticut
Location: Cobb 201
Time: 3:30pm

I present and discuss four or five universals drawn from across-linguistic study of comparative and superlative morphology. Special attention is given to three generalizations regarding root suppletion in the comparative degree of adjectives (good-better, bad-worse). These generalizations, I contend, have a variety consequences for morphology, semantics and perhaps syntax, particularly in the areas of lexical decomposition (at whatever level this obtains) and the formal treatment of suppletion vs. irregularity. Although comparative suppletion is rare (though attested) outside of Indo-European, and although the data sample is small within any one language, the generalizations over the total data set are surprisingly robust. Two generalizations are given here:

The Comparative-Superlative Generalization:

If the comparative degree of an adjective is built on a suppletive root/stem, then the superlative is also suppletive. The superlative may use the same root as the comparative, or may be further suppletive, but will not use the basic adjectival root. Thus the schema in (1), where A, B, C refer to phonologically unrelated roots.

(1) A - A - A completely regular: short, short-er, short-est
A - B - B suppletive: bad, worse, worst
A - B - C doubly suppletive: Latin ‘good’: bonus - melior -optimus
A - B - A *unattested* * bad - worse - baddest

I argue that this generalization favours analyses in which the superlative is not merely related to the comparative (e.g., both involve degree operators), but is rather _derived_from_ the comparative: [[[SHORT]-ER]-(ES)T]. Put somewhat more contentiously, I argue (with a qualification) that UG excludes a morpheme “-EST” (Superlative) that attaches directly to adjectival roots.

The Comparative-Change-of-State Generalization:

If the comparative degree of an adjective is built on a suppletive root, then a derived change-of-state verb (inchoative or causative) will also be suppletive. The verb may use the same root as the comparative (bad - worse - worsen; bonus -melior - meliorare), or may be further suppletive, but will not use the basic adjectival root.

By parity of reasoning to the first section, I must conclude (contra Dowty and others) that change-of-state verbs always include the comparative at some level of representation (cf. Kennnedy & Levin). I will defend this view against a variety of possible objections and examine apparent counter-examples.

Getting ‘Better’: On Comparative Suppletion and Related Topics

Jonathan Bobaljik
University of Connecticut

I present and discuss four or five universals drawn from across-linguistic study of comparative and superlative morphology. Special attention is given to three generalizations regarding root suppletion in the comparative degree of adjectives (good-better, bad-worse). These generalizations, I contend, have a variety consequences for morphology, semantics and perhaps syntax, particularly in the areas of lexical decomposition (at whatever level this obtains) and the formal treatment of suppletion vs. irregularity. Although comparative suppletion is rare (though attested) outside of Indo-European, and although the data sample is small within any one language, the generalizations over the total data set are surprisingly robust. Two generalizations are given here:

The Comparative-Superlative Generalization:

If the comparative degree of an adjective is built on a suppletive root/stem, then the superlative is also suppletive. The superlative may use the same root as the comparative, or may be further suppletive, but will not use the basic adjectival root. Thus the schema in (1), where A, B, C refer to phonologically unrelated roots.

(1) A - A - A completely regular: short, short-er, short-est
A - B - B suppletive: bad, worse, worst
A - B - C doubly suppletive: Latin ‘good’: bonus - melior -optimus
A - B - A *unattested* * bad - worse - baddest

I argue that this generalization favours analyses in which the superlative is not merely related to the comparative (e.g., both involve degree operators), but is rather _derived_from_ the comparative: [[[SHORT]-ER]-(ES)T]. Put somewhat more contentiously, I argue (with a qualification) that UG excludes a morpheme “-EST” (Superlative) that attaches directly to adjectival roots.

The Comparative-Change-of-State Generalization:

If the comparative degree of an adjective is built on a suppletive root, then a derived change-of-state verb (inchoative or causative) will also be suppletive. The verb may use the same root as the comparative (bad - worse - worsen; bonus -melior - meliorare), or may be further suppletive, but will not use the basic adjectival root.

By parity of reasoning to the first section, I must conclude (contra Dowty and others) that change-of-state verbs always include the comparative at some level of representation (cf. Kennnedy & Levin). I will defend this view against a variety of possible objections and examine apparent counter-examples.

Different alternatives for topics and foci: Evidence from indefinites and multiple wh

Stefan Hinterwimmer (joint work with Sophie Repp)
Zentrum fur Allgemeine Sprachwisenschaft, Berlin

Thursday May 15, 2008
Cobb 201, 3.30- 5 pm

In gapping, topical indefinites as well as wh-phrases can contrast with surface-identical antecedents if the contrast involved is the first of the two (or more) contrast pairs in the gapping coordination. This is not possible for most other types of expressions. We argue that both topical indefinites and wh-phrases introduce a discourse referent with a fixed address, on the basis of which referents introduced by surface-identical expressions can be contrasted. For the indefinites, we argue that the first contrast pair is a pair of contrastive topics which can, at the same time, be a pair of aboutness topics. These introduce individual addresses (Reinhart 1981). For wh-phrases we follow the assumption that they introduce discourse referents by presupposition. Multiple wh-interrogatives then introduce functions by presup po sition whose domain is provided by the first wh-phrase. The function is specified by giving its extension, i.e. the respective pair-list.

University of Chicago, Linguistics Colloquium

Syntactic phases and Codeswitching

Kay-Eduardo González-Vilbazo and Luis López
University of Illinois at Chicago (UIC)

Thursday, May 1 2008
Cobb 201

Abstract
Since Chomsky (1995) there has been ample debate on what exactly the role of little v is (see for instance Kratzer 1996, Marantz 1997 for two early proposals). After Chomsky (2000) and the development of phases as a theoretical construct, the question of little v’s role has become even more complex. This presentation aims to show that the linguistic competence of bilingual code-switchers provides a rich data base to test the value of the little v hypothesis. That is because speakers can switch between a lexical expression of little v and its complement VP, allowing us to extricate their respective contributions to the make-up of the sentence.

The grammar of bilingual code-switchers allows for a structure consisting of a light verb in one language (L1) followed by the main predicate with its arguments in the other (L2). This is exemplified in (1), with L1 Spanish and L2 German. The striking fact is the following: although the constituents of a are fully German in structure, the constituent order, prosodic structure and expression of focus/background of a itself follow the rules and restrictions of Spanish.

(1) Juan ha hecho [a verkaufen die Bücher].
Juan has done sell the books
‘Juan has sold the books.’

Juan ha hecho –> L1, Spanish

Verkaufen die Bücher –> L2, German

We find that little v is directly involved in at least three linguistic properties: linearization of the lexical verb V and its complements, the prosodic structure of VP in neutral contexts and the expression of Focus/Background structure. Thus, features of little v determine (at least) the outcome of the mapping between syntax and PF and syntax and information structure.

There will be a special colloquium next Friday from Enoch Aboh of the University of Amsterdam and MIT. The title of the talk is “A Typology of Adpositions” and you can find the abstract here:

http://clml.uchicago.edu/svn/filedetails.php?repname=CLML+Repository&path=%2FPapers%2FPrePostpositionsChicago.pdf

The talk will be held at the normal colloquium time in the normal colloquium location (3:30 in the CSL)

Modularity in Morphology: The Case of Basque Finite Auxiliaries

Karlos Arregi
University of Illinois, Urbana-Champaign

Thur, Feb 21 2008 3:30-5:00pm, Cobb 201

Many modern theories of morphology are highly modular: morphological phenomena with different properties are accounted for by separate modules of grammar. In this talk, I argue for a particular view of the modularity of morphology based on examination of Basque finite auxiliaries, which have a complex system of clitics that cross-reference ergative, absolutive and dative arguments in the sentence. I argue that a principled analysis of several generalizations regarding these clitics must involve a theory of word-formation with separate modules that have their own well-formedness principles and repair operations. Special attention will be given to verbal forms where these requirements of the separate modules conflict with each other. It will be shown that these conflicts are resolved in a manner that is best accounted for by a theory where the modules involved in word formation are derivationally ordered.

On the parametric variation of case and agreement: implications for (non)-configurationality

Vita Markman
Simon Fraser University

Henry Hinds [5734 S. Ellis Ave.] Room 101.

Tues Feb 19, 2008 3:30-5:00pm

In this talk I will argue that case and agreement features are subject to parametric variation and explore the consequences of this claim with a particular attention to word order. Departing from the view that case and agreement are present in the syntax of every language, but may not be overtly realized (Rouveret and Vergnaud 1980; Chomsky 1981, 1995, 2000, 2001; Harley 1995; Bittner and Hale 1996; Sigurdsson 2003), I will argue that languages can choose to have case features, agreement features, some combination of the two or none at all. The main focus of the talk will be on languages that have agreement features but no case. Specifically, I will demonstrate that languages without case features, but with agreement features will be non-configurational. These include Mohawk, Kinande, and Chichewa. In contrast, languages with case features may allow but not require NP dislocation in the presence of agreement. These are all of the Indo-European languages, Japanese, and Nahuatl.

In addition to addressing the effects of the parametric variation in case and agreement on word order, I will also address a number of other syntactic phenomena that pose a problem for the ‘universal’ approach to case and agreement and are better understood if these features are taken to vary parametrically.

Sometimes Syntax is Syntax. Sometimes Syntax is Phonology

Jason Kandybowicz
Swarthmore College

Thursday, February 14, 2008, 3:30-5:00pm
Cobb 201

In recent years, the field of syntax has seen a shift toward explorations and explanations of syntactic phenomena cast in terms of the interfacing sub-systems of grammar; namely, the phonological and semantic components. This modus operandi necessitates a broader knowledge base than was previously thought necessary. It also entails that the more rigorous analyses in this vein will likely come from those who study languages holistically. Yet curiously enough, holism is far from being the battle cry in today’s interface-driven syntactic frameworks. In this talk, I advance an argument for linguistic holism on the basis of two case studies drawn from the Nupe language, a Benue Congo language spoken in south central Nigeria.

The first case study deals with the language’s restriction on extraction from perfect clauses. The second case study is similar in that it too deals with an extraction restriction. In this case, the restriction involves the prohibition of embedded subject extraction across a complementizer – the so-called Comp-trace effect. Although the phenomena investigated in both case studies have been traditionally referred to as “syntactic” in both the Nupe literature and in the generative literature more broadly, I show that the former is truly syntactic in the narrow sense, while the latter is more phonological in nature. In this respect, then, it is difficult to know in advance of analysis whether a purported syntactic phenomenon is truly syntactic after all. Thus, in light of situations like these, holistic approaches to language take on an elevated level of importance. The talk also addresses a number of theoretical issues raised by the core empirical problems of each case study, including, but not limited to, the syntax-phonology interface.

Northwestern University Department of Linguistics Colloquium presents

On the Role of Syntax on the Interpretation of Elided Reflexives

Jeffrey Runner
University of Rochester

The main explanations for the exceptional behavior of reflexives in “representational NPs” (RNPs), e.g., ‘a picture of herself’, rely on syntactic or argument structure (Chomsky, 1986; Davies & Dubinsky, 2003; Pollard & Sag, 1992; Reinhart & Reuland, 1993). “Reference transfer” (RT) allows reference to a representation of a person by that person’s name, e.g., referring to a statue of Ringo Starr as ‘Ringo Starr’ (Jackendoff, 1992). Like RNP reflexives (Grodzinsky & Reinhart, 1993), RT reflexives may receive coreferential interpretations when elided (Lidz, 2001). Here I present evidence from collaborative work with Micah Goldwater (UT Austin) of two scene verification experiments and two “visual world” eye-tracking experiments suggesting that it may be the representational use of RNP reflexives- and not (just) the syntactic/argument structure- that allows for their exceptional behavior. Interesting differences are found between the two sets of experiments, which can also shed light on the approaches to ellipsis interpretation discussed by Kehler (2000) and Frazier & Clifton (2006).

Friday, February 15, 2008
3:30 p.m.
Chambers Hall (600 Foster Street), Lower Classroom Level

Zenzi Griffin (Georgia Institute of Technology) will speak at the Department of Psychology colloquium on January 17, Thursday, at 4pm in Rosenwald 011. Her talk is titled, “How speakers’ eye movements reflect spoken language generation“.

Gradient phonological generalizations in speech processing

Mirjam Ernestus
Radboud University Nijmegen &
Max Planck Institute for Psycholinguistics

Thursday Jan. 10, 2008
Cobb 201, 3.30-5 pm

Abstract
Several studies have shown that speakers are sensitive to the absolute and statistically gradient phonological patterns in their mental lexicon. Participants prefer words conforming to these patterns, and they are slower in producing morphologically regular word forms violating the patterns. In this talk I will discuss two series of experiments that further investigate the role of gradient patterns in speech processing. The first series suggest that generalizations based on intraparadigmatic relations, between the forms of single word, have a stronger effect than those based on interparadigmatic relations, between the same types of forms of different words. The second series of experiments shows that phonologically gradient patterns affect also speech comprehension, even when listeners are focusing on content instead of form. This shows that gradient generalizations play a role in everyday language processing.

The Department of Linguistics Colloquium Series presents

Andrea D. Sims

Northwestern University

“When synchronic motivation disappears: On probabilities, paradigms, and processes of lexicalization”

Thursday, November 1, 2007

Cobb 201, 3:30pm

ABSTRACT

In this paper I explore some of the conditions under which inflectional structures become lexicalized, and why they persist once established in the lexicon. I present a case study of Russian verbs which are defective in the 1sg non-past (e.g., *pobezhu ‘I will win’). Starting with Halle (1973), these verbs have presented a 30+-year mystery for two reasons. First, there is no apparent synchronic motivation for these verbs being defective, yet the paradigmatic gaps are not filled by productive inflection. Second, it is generally accepted that the gaps have been lexicalized, but theories of lexicalization fail to explain why this happened (Brinton and Traugott 2005).

Based on experimental evidence and agent-based simulation, I argue that the lexicalization and continued existence of defectiveness among Russian verbs can be understood as a paradigm effect. Specifically, gaps represent an interaction between morphosyntactic probability distributions and morphophonological coherence. The core tenets of my proposal draw from existing literature: two-level paradigmatic structure (Stump 2006), sensitivity to probability distributions across the lexicon (Kemps et al. 2005), and a Bayesian learning mechanism (cf. Regier and Gahl 2004). However, this paper represents one of the first attempts to synthesize these disparate lines of research to explain inflectional change.

The results argue for a more extensive role of paradigmatic structure in promoting lexicalization than has previously been recognized (Aski 1995, Maiden 2004). It also shows that lexicalization is controlled by frequency in subtle ways, beyond the common observation that high frequency items are more likely to be entrenched.

The Department of Comparative Human Development Colloquium Series presents

JOHN SEARLE

Professor, Department of Philosophy
University of California, Berkeley

“Language and Social Ontology”

Tuesday, November 6th, 2007, 4:30pm
Social Sciences 122

ABSTRACT
In this talk I will show how all of human institutional reality is created and maintained by a single logico-linguistic process. The work is a continuation of the ideas in “The Construction of Social Reality.”

The Department of Linguistics at the University of Chicago presents

MARINA TERKOURAFI
University of Illinois Urbana Champaign

Cooperation revisited and renewed

Thursday, October 25, 2007
Cobb 201, 3.30-5 pm

Since (at least) Bach and Harnish (1979), several attempts have been made to clarify what might be “the accepted purpose…of the talk exchange” mentioned by Grice (1975) in the definition of the Co-operative Principle, and, more generally, the nature (and the limits) of cooperation enjoined by the CP itself. Following a brief critique of previous proposals, I suggest the notion of face, adapted from the anthropological linguistics literature, as an omni-relevant ‘accepted purpose’ on which more specific purposes may be superimposed. The omni-relevance of face follows from two properties: its biological grounding in the dimension of approach vs. withdrawal, and its intentionality (i.e. aboutness). I am thus advocating a revised notion of face, which operates on two levels. As a second-order notion, ‘Face2’ is universal (qua biologically grounded), while, at the same time, it is uniquely human and irreducibly relational (qua intentional, i.e. directed at an Other defined as distinct from Self). This second-order notion of face is, nevertheless, no more than a convenient methodological abstraction. Face as a first-order notion is the only one that has any psychological reality for speakers, and is fleshed out in particular socio-historical circumstances, resulting in a multiplicity of ‘Face1’s simultaneously operating in interaction. Contrary to previous approaches that viewed considerations of face as “principled reasons for deviation [from the CP]” (Brown & Levinson 1987: 5), face in this novel dual conceptualization is claimed to operate before the CP, regulating the generation of implicatures by prompting recursive application of the maxims until one is satisfied that one knows (with some degree of certainty) how one stands in relation to one’s interlocutor(s). In other words, an omni-relevant purpose of interaction is ascertaining one’s standing in relation to one’s interlocutor(s)—in the sense of determining whether one’s face has been constituted or threatened—and this purpose determines a (context-dependent, hence variable) cut-off point for the inferential process. I illustrate these suggestions by re-analyzing some classical examples from the pragmatics literature, demonstrating how this expanded understanding of face and of the CP can account for behaviors ranging from over-cooperation (altruistic behavior) to outright conflict.

The Department of Comparative Human Development Colloquium Series presents

JOHN SEARLE

Professor, Department of Philosophy
University of California, Berkeley

“Language and Social Ontology”

Tuesday, November 6th, 2007, 4:30pm
Social Sciences 122

ABSTRACT
In this talk I will show how all of human institutional reality is created and maintained by a single logico-linguistic process. The work is a continuation of the ideas in “The Construction of Social Reality.”

The Department of Linguistics presents

Jason Stanley

Rutgers University

Knowing How in Romance

October 18, Thursday, 3pm, Cobb 201

Workshop on Semantics and Philosophy of Language presents

Jason Stanley

Rutgers University

Sly Pete: Indicative Conditionals and Context Dependence

October 19, Friday, 11am - 1pm, Wieboldt 111

William Croft

University of New Mexico

When is a modifier a modifier? A Radical Construction Grammar approach

Thursday, October 11, 2007, 3:30-5pm, Cobb 201

The topic of this year’s Workshop on Semantics and Philosophy of Language, organized by Chris Kennedy and Anastasia Giannakidou this year, along with Josef Stern from Philosophy, is “compositionaltiy”. External speakers this fall include Zoltan Szabò (Oct 12), Jason Stanley (Oct 19), and Polly Jacobson (Nov 9). For more details, visit the workshop website at http://semantics.uchicago.edu/workshops/spl/.

The talks that Jason Stanley was scheduled to give next week — a linguistics colloquium on Thursday and a semantics/philosophy workshop talk on Friday — have been postponed until fall quarter, as Jason will not be able to make it to Chicago next week.

The workshop on semantics and philosophy of language is pleased to announce a talk this Friday, May 25, by Chris Barker from NYU. The talk will take place 11-1 in Classics 47. The title of his talk is “The case against E-type donkey pronouns”. Some background reading can be downloaded from the workshop website:

http://semantics.uchicago.edu/workshops/spl/barker.html

Jerry Sadock will give a colloquium on Thursday, May 17, in Harper 130 at 3:30pm. The title of his talk is “English Tense and the Auxiliary System.”

Teresa Statterfield‘s colloquium on Thursday, May 10 has been postponed. Please check the colloquia page for updates.