Department of Linguistics

According to phonetically-based phonological frameworks, functional constraints such as perceptual distinctiveness play a central role in shaping phonological behaviors (Boersma, 1998; Hayes et. al, 2004). This view is challenged by evidence of phonetically unnatural patterns active in synchronic phonological grammars (Anderson, 1981; Hyman, 2001). I consider arguments for the phonetic grounding of phonological features in Vietnamese tone, where it has been argued that, despite dialectal differences in the phonetics of tone production, phonetically grounded tone features are shared across dialects (Pham, 2001, 2003). From the results of a cross-dialectal perception study, I argue that the features relevant for the perception of tones no longer correspond to their phonologically active counterparts in any straightforward way, either within or between dialects. This result is discussed in terms of its implication for the notion of phonetically grounded phonological constraints, as well as for the relationship between subphonemic and categorical levels of linguistic structure.

Phonological contrasts make crucial reference to features, which are in turn signaled by various subphonemic cues. When the balance of cues changes sufficiently such that a previously intrinsic cue becomes extrinsic, this cue is said to have become phonologized (Hyman, 1976). A considerable body of recent work has focused on the extent to which recurrent patterns of phonologization should be attributed to channel effects (such as misperception) or to analytic biases (such as UG). While this approach highlights the potential differences (or lack thereof) in evidence favoring one mode of explanation over another, it is silent on the issue of precisely how these biases influence ongoing sound change. In this talk, I present a framework for investigating the relative contribution of bias in phonologization by modeling phonetic contrasts using Gaussian mixtures (Nearey & Hogan, 1986). I argue that phonologization is the result of an adaptive enhancement strategy, driven by loss of contrast precision, that optimizes both listener- and speaker-oriented constraints (Lindblom et al., 1995). Subphonemic cues are targeted and enhanced in a probabilistic fashion, proportional to their informativeness and the precision of the contrast they help to cue. I will present the details of this model and illustrate with an account of the ongoing phonologization of pitch in Seoul Korean (Kang & Guion, 2008).

I will describe on-going work that examines the division between lexical- and discourse-level processes in the interpretation of quantified and co-referential expressions. Part 1 examines how different forms of a quantified expression can highlight distinct sets within discourse. Participants’ responses in a passage-completion task and a self-paced reading task indicate that while positive forms like ‘some’ highlight the specified amount (the reference set), negative forms like ‘only some’ are ambiguous between this quantity and the remaining amount (the complement set). Part 2 tests whether this discourse ambiguity introduced in the negative quantified context had an immediate or delayed influence on co-referential processing using an eye-tracking while reading paradigm. At the lexical level, early eye-movements indicate immediate facilitation during reading of repeated anaphoric expressions. However, relative to this benchmark, the influence of context on the processing of referential ambiguity at the discourse level is much slower. Finally, part 3 examines how lexical and discourse processes are distinguished in the brain by comparing the interpretation of new and repeated expressions using event-related potentials. New expressions generate greater negativity during the 250ms-500ms window (N400) while infelicitous co-referential repeated expressions generate greater positivity during the 500ms-800ms window (P600). Altogether these results suggest a robust division between lexical and discourse-level processes during language comprehension.

Language acquisition has traditionally been conceived of as a matter of acquiring representations, such as words or syntactic rules. However, the young learner must also develop the ability to coordinate these representations in order to comprehend utterances in real-time. Do children consistently generate the same kinds of interpretations as adults do during comprehension? In cases where they do not, what is the nature of this developmental difference and what does it reveal about the connections between various linguistic representations? I will introduce three lines of work that investigate these issues by comparing patterns of sentence processing in five-year-olds and adults. Part 1 examines whether cascaded processing in word recognition is a basic architectural feature of the lexicon or one that is based on extensive linguistic experience. Part 2 examines how morphosyntactic markers which identify thematic roles influence interpretation of passive sentences in Mandarin-speaking children. Part 3 examines whether children can generate post-semantic inferences to capture a speaker’s intent. Altogether these results suggest that like adults, children exhibit incremental processing of language. However, unlike adults, children often fail to inhibit/revise initial interpretations. More broadly, this work highlights ways in which viewing language as a series of linked representations and comprehension as a process sheds light on the developmental trajectory of children’s interpretations.

There are a growing a number of studies investigating "illusions of grammaticality" across different long-distance dependencies. Across this work, it appears that some dependencies (e.g. agreement and negative polarity items in Pearlmutter, Garnsey & Bock 1999; Drenhaus, Saddy & Frisch 2005) are highly susceptible to these illusions, while others appear to remain more grammatically faithful (e.g. reflexives in Clifton, Frazier & Deevy 1999; Sturt 2003). Due to a number of possible differences between the relevant studies, however, direct comparison is difficult to do. This session describes work in progress that directly compares reflexive and agreement dependencies to contrast the effect of these interference effects across dependency type. These dependencies provide a crucial test for the role of structure-independent feature-matching operations in processing, as they require retrieval of information at the same structural address. Initial results support the hypothesis that reflexives in general are not as susceptible to interference as agreement. This result, if true, provides further evidence that feature-matching is structurally constrained. However, proving that reflexives show no interference at all represents a difficult statistical problem; it requires the researcher to demonstrate the truth of the null hypothesis, which is not generally possible with traditional methods of hypothesis testing. In the current experiments, proving a null result carries significant theoretical purchase. In order to begin to make this argument, I will present an alternative Bayesian method of data analysis (Gallistel 2009) that allows inference about the existence of null effects.

The field of sentence processing is in the middle of a revolution in the way people view the relation between domain-general working memory processes and parsing routines, driven by novel experimental data (e.g. McElree, Foraker & Dyer 2003; Van Dyke 2007) and successful computational models (Lewis & Vasishth 2005). By tying together insights from psychology and computer science, researchers are now in a position to make specific claims about the computational character of the memory systems that drive parsing. In this talk, I will explore the contribution of linguistic structure to this line of research by presenting two case studies that examine the interaction between linguistic representation and working memory processes. In the first part of the talk, I examine the role of linguistic structure in gating memory access by examining the Mandarin Chinese long-distance reflexive ziji. Long-distance reflexives are an interesting test case because they require the use of structural information to resolve their antecedent, and can have potentially unbounded search spaces. I present speed-accuracy trade-off data that indicates the use of a hierarchically structured search process in resolving ziji's antecedent-anaphor dependency, along with data from event-related potentials that suggests possible algorithmic implementations of this search. In the second part of the talk, I examine a subclass of interference effects known as 'partial cue overlap' effects. By comparing negative polarity items (NPIs) and reflexives in English, I will argue that the locus of these effects is in processing errors that are particular to NPI dependencies, rather than an architectural fact about the parser. These results suggest that memory search strategies are sensitive to linguistic structure in surprising ways during online processing, a finding that in turn suggests new ways of exploring the interface between working memory and linguistic structure-building.

Many researchers have proposed that phonological processes recruit listed allomorphs in order to satisfy constraints on surface structure. Such proposals can offer a unified treatment of both suppletive and regular allomorphs, but at a cost: the phonological alternations that characterize regular allomorphs are re-cast as listed morphological alternations. In a study of French adjectival liaison, I seek evidence as to whether this cost is justified. Liaison offers a good test case because it recruits both suppletive allomorphs (bel[le] ‘beautiful-FEM’ versus beau ‘beautiful-MASC’) and regular allomorphs (parfaite ‘perfect-FEM’ versus parfait ‘perfect-MASC’) in order to satisfy a constraint against vowel hiatus (Perlmutter 1996, Tranel 1996, Steriade 1999). Preliminary results from a word recognition task focusing on regular liaison adjectives indicate that listeners treat these words as if they involve a change in morphological gender, rather than the insertion of phonological material. The implication is that even phonologically regular allomorphs bear evidence of distinctly morphological origins.

Although non-local dependencies are common within and across sentences, they are relatively rare within words, limited primarily to cases of vowel and consonant harmony. As a result, word structure often seems to require a strictly local analysis distinct from other levels of language. In three studies focusing on lexically contrastive pairs such as bite versus bide, I pursue the hypothesis that non-locality does occur regularly within words, at the fine-grained level. The bite-bide pairs make a good test case because their diphthongal vowels offer both a local and a non-local target for the vowel lengthening triggered by voiced final consonants.

The production study demonstrates that speakers implement the bite-bide contrast with non-local dependencies which are absent from other alternations that also condition vowel lengthening, such as phrasal position and speech rate. The two perception studies demonstrate that listeners can use this dependency to identify bite versus bide. These results provide evidence that non-local dependencies do occur within the fine-grained structure of the word, and furthermore that such dependencies act as a unique signature for lexical contrast. An appropriate analysis of word structure therefore must permit, and motivate, non-locality.

Spanish and English only display “long-distance” wh-questions such as A quién piensas que ha conocido Marsias? (Who do you think Marsias has met?) while languages such as German allow “scope” wh-questions such as Was glaubst du wen Samir getroffen hat? (*What do you think who Samir has met?) as well as “copy” wh-questions such as Wen glaubst du wen Samir getroffen hat? (*Who do you think who Samir has met?). In spite of the fact that neither “scope” nor “copy” wh-questions are grammatical in English, both have been shown to occur in the adult L2 English grammars of speakers whose L1s do not exhibit them. This presence has been accounted for as evidence that adult L2 learners have access to Universal Grammar (UG) principles. Namely, these learners produce and accept these constructions because even though they are neither available in the input nor in their L1, they exist in other natural languages. However, since we do not know whether non-native grammars depict all the possible constructions available in natural languages, the reason why “scope” and “copy” constructions occur calls for a more compelling explanation. First, since “scope” and “copy” constructions have not been found in spontaneous production, one could conclude that their presence is induced by the actual experimental tasks. Second, these constructions seem to occur in the non-native grammars of speakers with a low level of proficiency and never at rates higher than 15%. In fact, the “target” constructions are always the preferred choice and, while the rate of the “scope” construction could reach 15%, the rate of the “copy” construction is always lower. This could suggest that they are an interlanguage phenomenon due to either a processing need (i.e. every C position has to be filled with a wh-Comp) or to a misanalysis of the complementizer or the actual wh-word. It could also suggest that there is a universal hierarchy according to which “long-distance” would be the default option and “scope” would come in second place. Actually, it comes as no surprise that “scope” and “copy” have a different status in the English non-native grammars because there is no consensus in the literature as to whether or not the two constructions should be analyzed in similar terms. In fact, it has been argued that in “scope” constructions the matrix wh-word does not move from the embedded clause but represents an independent question from the embedded one.

Based on the above, we have carried out a study intended to investigate whether: (i) both “scope” and “copy” wh-questions occur in L2 developing grammars other than English where neither the L1 nor the target grammar exhibit them; (ii) input and transfer determine the status of these constructions in non-native grammars (iii) the “target” > “scope” > “copy” pattern of preference attested in L2 English developing grammars constitutes evidence for the existence of a universal availability hierarchy and (iii) “scope” and “copy” have a different status in both native and non-native Spanish and native and non-native German.

The results of a grammaticality judgments task administered to two groups of adult English and French speakers and two groups of adult Czech speakers learning German and Spanish as foreign languages, as well as to two control groups of native German and Spanish speakers show that: (i) the presence of these constructions in non-native Spanish—and therefore in other non-native languages where neither the L1 nor the target grammar exhibits them—may be task-induced; (ii) input and mainly transfer play a clearer role in the case of “scope” than in the case of “copy” wh-questions; (iii) the hiearchy does not hold; and (iv) there is no evidence that “scope” wh-questions are interpreted as instances of two different questions.