UNDERSTANDING LANGUAGE:

FORTY YEARS DOWN THE GARDEN PATH

ABSTRACTS (Download in pdf)

Where do Linguistic Universals Come From

T.G. Bever, University of Arizona, USA.
 

Language has been argued to be a unique skill as a function of a special set of underlying evolved capacities. Thus, it is modular, has a unique neurological organization, has a ‘critical age’ for acquisition, is largely free of the effects of induction, has no structures specifically to motivate its acquisition. I will review arguments and facts about such phenomena to show that they may be either false or not unique to language. The computational architecture and processes in language remain, however, as a unique assemblage of neurological and computational constraints.

Neurotypology: Modelling cross-linguistic similarities and differences in the neurocognition of language comprehension

Ina Bornkessel-Schlesewsky, University of Marburg, Germany

Linguistic diversity poses a formidable challenge to the quest for understanding the human language processing architecture: how can the existence of over 6000 languages be reconciled with the assumption of a single (neuro-)cognitive language processing system? This paper reviews a new approach to this issue, entitled "neurotypology". By comparing and contrasting neurophysiological processing signatures for languages from different language families and with different typological characteristics, neurotypological research aims to establish cross-linguistic generalisations in the neurocognition of language as well as to identify dimensions of variation. I will argue that the generalisations identified in this way are promising candidates for "cognitive attractors", which serve to shape the structure of language(s) on the one hand and provide a new perspective on the structure of human cognition on the other. This proposal will be illustrated with reference to one such potential attractor, the notion of "actorhood", and one major cross-linguistic dimension of variation, sequence-driven vs. sequence-independent verb-argument linking. The claims for both domains are based on neurophysiological data from a range of languages, including Chechen, German, Hindi, Icelandic, Japanese, Mandarin Chinese, Tamil, and Turkish. From these data, I conclude that universals of language and language processing are intimately tied to the structure of human cognition and neurobiology and that, in addition to constituting a fascinating field of investigation in and of itself, a cross-linguistic approach to language processing could help us to gain a deeper understanding of how we think and how our brain works.

Mechanisms of agreement

Manuel Carreiras, Basque Center on Cognition, Brain and Language.

Agreement plays an important role in language comprehension especially in richly inflected languages such as Spanish. Person, number and gender, which can be a conceptual or formal property, are features for agreement. In this talk I will describe behavioral and neuroimaging evidence on the agreement processes in language comprehension with different features mainly in monolinguals, but also in late learners of Spanish that do not have formal gender in their L1.  The main addressed questions are a) how the processor solves agreement on-line when semantic or morphosyntactic features are available or absent; b) Is agreement computed similarly for all features; c) do second language learners compute agreement as in their first language; d) what agreement tell us about mechanisms for sentence processing.

Implicit learning in the language production system is revealed in speech errors

Gary Dell, University of Illinois

I'll first describe a general framework for psycholinguistic theorizing that I and several other psycholinguists have been working with over the last 10 years. This framework hypothesizes a particular relationship among processing (comprehension), production, and acquisition. Then, I'll illustrate this framework by summarizing experiments in which subjects implicitly learn artificial phonotactic constraints in the laboratory. One interesting aspect of the learning is that it expresses itself in the learner's speech errors. Another is that there seem to be different learning mechanisms for simple and complex constraints.  The challenge for research in which participants learn artificial patterns in the laboratory is that, thus far, the research has had only limited success connecting its findings to processing outside of the lab and particularly to language acquisition by children.

From Action to Language: Evidence and Speculations

Luciano Fadiga Italian Institute of Technology, U. Ferrara

In recent years we have empirically shown that the observation of actions performed by others automatically activates the observer's premotor cortex. These data, coming from electrophysiological recordings in monkeys (mirror neurons) and from TMS and brain imaging studies on humans, support the idea that this observation-evoked “motor resonance” could be the functional expression of a brain mechanism involved in understanding the actions of others. However, it might appear paradoxical that the motor centres, classically considered as the output stage of the brain, are involved in such high-level cognitive task. In the first part of my presentation I will discuss this issue by showing that mirror neurons are indeed part of a more general mechanism involving motor representations in perceptual processing. In the second part of my talk I will show that during speech listening TMS reveals an enhancement of excitability of the listener's tongue motor representations and that, according to some very new data, this motor involvement contributes to speech perception. Finally, I will discuss the involvement of Broca's area in action understanding, being this speech-related centre constantly activated by the observation of actions of others.

Changing perspectives on the functional role of some language regions in the brain

Yosef Grodzinsky, McGill University, USA

A functional anatomy of cognition attempts to superimpose cognitive operations onto brain pieces. A functional anatomy of language subsequently attempts to identify pieces of linguistic ability that reside in the language regions of the brain. A prominent tradition, established by Wernicke and Lichtheim, and promoted by Geschwind, views language production and comprehension, reading, writing and repetition as the elemental functional pieces, interconnected to form the language network. In the last quarter of the 20th century new perspectives emerged, that provided empirical arguments that cast doubt on this approach. Thus the perspective on the functional role of at least some of the language regions has changed, gravitating toward a view that pieces of combinatorial linguistic ability - as characterized by linguistic theory - are the ones that best align with the anatomy. I will discuss some recent developments, and present the latest arguments for a change in perspective in relation to Brodmann's celebrated cytoarchitectonic map, that has been reincarnated as the probabilistic atlas of the Jülich brain mapping project. The experimental results I will review come from comprehension experiments with highly complex sentences in several languages, conducted with patients suffering from traumatic aphasia, and with healthy adults in fMRI. I will use these to argue that the most relevant functional pieces of the anatomy of language are rather small, as certain combinatorial linguistic operations seem to align rather well with cytoarchitectonic borders.

Statistical Generalizations in Language Behaviors

William Idsardi, University of Maryland, USA

Bayes's Rule tells us that the probability of a hypothesis, h, given data, d, is proportional to the probability of the data given the hypothesis and the prior probability of the hypothesis, p(h|d) α p(d|h)•p(h). Bayes's Rule has a direct connection to the Analysis-by-Synthesis method proposed a half-century ago by Halle & Stevens (1962): p(h|d) is the analysis, p(d|h) is the synthesis (i.e. forward model) and the prior, p(h) represents the initial state of the hypothesis space (i.e. UG). The past decade has seen an enormous rise in the application of Bayesian methods to models of perception and learning. We will review some recent Bayesian models for speech sound perception and learning, and show how the Bayesian approach yields a model-comparison procedure that sheds important new insight into classic problems in phonology. We will concentrate on the power of the forward model in constraining solutions to be checked against the incoming data, elaborating an initial sketch into a full analysis, consistent with Reverse Hierarchy Theory (Ahissar & Hochstein 2004, Nahum, Nelken and Ahissar 2008, Shamma 2008).

Sonja A. Kotz, Max Planck Institute for Human Cognitive and Brain Sciences

Neural correlates of agreement: locality,  distance, and sentence complexity.

Verb form provides relevant grammatical properties of the subject noun phrase, such as number and person in many languages. As such subject-verb agreement aids the assignment of thematic roles to the elements of a sentence in language comprehension. Given the fact that subject and verb can occur at variable distances from each other and can be separated by elements that carry interfering syntactic information, it is an interesting question how the human brain accomplishes these challenging tasks. In my talk I will provide temporal (ERP) and spatial (fMRI) evidence from phrase and sentence processing studies addressing locality, distance, and complexity.

Charles Lin, Indiana University

Down the head-final garden path: Understanding the processing asymmetries of head-final relative clauses

The processing asymmetry between subject- and object-extracted relative clauses (RCs) reported across languages have been crucial data that various sentence-processing theories attempted to account for, including, for instance, theories that focus on the structural position of subject and object gaps (Keenan & Comrie, 1977), the working memory consumed in maintaining and integrating dependencies (Gibson, 1998; Lewis et al., 2006), and the dominant syntactic and semantic patterns in a language (Bever, 1970; MacDonald & Christiansen, 2002). Recent studies on the processing of head-final relative clauses (Chinese, Japanese, and Korean; e.g., Hsiao & Gibson, 2003; Hsu et al., 2006; Kwon et al., 2010; Lin & Bever, 2006, 2007; Ueno & Garnsey, 2008, Wu & Gibson, 2009, among others) have focused on teasing apart predictions of these theories. This talk examines these recent advancements of head-final RC comprehension. I will propose the possible reasons for the inconsistent findings in the literature and elaborate on an account that integrates the processing differences across head-final relative clauses in East Asian languages. The factors that will be discussed include the role of syntactic and information structure, the effect of RC-inducing contexts, frequent patterns in the language, and the issue of locality.

The Production Basis of Language Comprehension:  Evidence from Relative Clauses

Maryellen MacDonald, University of Wisconsin-Madison, USA

In the Production-Distribution-Comprehension (PDC) account of language processing, production constraints cause speakers to prefer certain structures and lexical-structural pairings over others, leading to broad distributional regularities in the language. Language users implicitly learn these patterns and apply them as probabilistic constraints to interpreting new input, as in constraint-based models of comprehension. Thus according to the PDC, comprehension preferences need not reflect particular properties of the comprehension architecture by may instead reflect biases developed from learning over patterns in the input, which themselves stem from production constraints. This talk will review production, learning, and comprehension studies testing the PDC, using relative clause materials.  Our studies in several Indo-European and Asian languages investigate why speakers avoid object relative clauses ("The reporter that the senator attacked…") in some environments and frequently produce them in others, and the extent to which these patterns are robust cross-linguistically over wide variation in word order, case marking, head direction, and other parameters.  Our learning studies investigate the extent to which adult language users can rapidly refine their knowledge of the environments in which object relatives are vs. aren't used in various environments, and our comprehension studies show that patterns of reading times in object relatives and other relative clause structures can be traced to the patterns of production and learning we have identified.  These results will be used to contrast the PDC approach with other accounts of relative clause comprehension difficulty, and challenges for the approach will be discussed.

Grammatical Illusions: Where you see them, where you don't

Colin Phillips, University of Maryland, USA

Grammatical constraints impose diverse requirements on the relations between words and phrases in a sentence. Research on the on-line implementation of grammatical constraints reveals a strikingly uneven profile. Speakers shows impressive accuracy in the application of some rather complex constraints, but make many errors in the implementation of some relatively simple constraints. Just as the study of optical illusions has played an important role in the study of visual perception, this highly selective vulnerability to interference and ‘grammatical illusions’ in the language comprehension system provides a valuable tool for understanding how speakers mentally encode structured linguistic representations and how they navigate those representations in real time.

Comprehension, production and linearization in a new evolutionary perspective

Massimo Piattelli Palmarini, University of Arizona, USA

In a Minimalist perspective, Narrow Syntax (NS) is conjectured to be the optimal solution to the problem of the interfaces (the sensory-motor-perceptual one and the interpretive one). The optimisation, however, seems to be such mainly for the second interface, while the first appears to have to cope with the “strange” requirements of the articulatory-perceptual apparatus(es). The existence of un-interpretable features and the requirement that they be locally checked and then deleted before reaching the interpretive interface is the paradigmatic case, and poses several problems about the optimality of the overall system (the Broad Faculty of Language).   Having, at long last, ruled out the canonical neo-Darwinian adaptationist reconstruction of the evolution of language, we have to indicate new approaches to the origins of language design.  The ubiquity of transposable elements, recently forcefully stressed by Carl Woese as the main evolutionary process for the longest period of evolution, suggests that there may have been insertions of genetic material into pre-existing structures (NS as a viral infection, forced to negotiate with pre-existing systems).  The linearization requirement imposes to flatten highly elaborate lexico-syntactic hierarchies onto a uni-dimensional time line (more or less along the frame of the Linear Correspondence Axiom LCA of Kayne 1994). Derivations proceed right-to-left, inside-out, while production and parsing proceed left-to-right and outside-in. Garden-path, restart, reanalysis and remnant phenomena may well be explained as the result of this conflict. An evolutionary (non-adaptationist) approach might shed some light on these phenomena

Syntax, semantics and pragmatics in incremental interpretation
Edward Stabler, UCLA, USA

Minimalist grammars and certain other "mildly context sensitive" formalisms can be used directly in probabilistic beam parsing, computing grammatical analyses simply and exactly, even when they involve relations like remnant movement that pose difficulties for naive and informal theories. Statistical, semantic and pragmatic influences can be accomodated in a way that impacts the analysis on a word-by-word basis. This sets the stage for the careful consideration of a range of processing models that can be grammatically sophisticated and quantitatively precise. Considering polarity-sensitive items as an example, this paper shows how structural and non-structural factors can interact in the recognition of scopal relations. The separation of syntax from pragmatics and other factors allows the grammar to have a basic structural property that makes these simple models possible.

On the Cognitive Basis of Linguistic Structures: Themes that have endured.

Michael Tanenhaus, University of Rochester, USA

Early work on ambiguity resolution focused primarily on the question of whether or not ambiguity complicated processing. One of the many contributions of Bever (1970) was to highlight both how temporary ambiguity was a problem in and of itself and how it could be used as a domain for examining language comprehension. Enormous progress came from combining studies of ambiguity resolution with real-time processing measures.  During the last decade my collaborators and I have been using referential ambiguity as a tool for examining real-time language processing during interactive conversation, beginning to merge two traditions that separated in the early days of psycholinguistics, what Clark (1992) dubbed the “language-as-action” and “language as product” traditions. I’ll (a) review recent work on how interlocutors construct and use referential domains; (b) argue that a fundamental puzzle in language processing has shifted from “how can language processing be so fast and seemingly effortless” to “how can construction and use of contextual domains be so fast and seemingly effortless”; and (c) draw some analogies to the growing literature on vision in natural tasks.