theme Following the PhD defense of Hartmut Fitz on Neural Syntax, a public symposium will be held on neural network approaches to processing natural language. Participation is free and registration is not required. The event is sponsored by the Institute for Logic, Language, and Computation.
time Wednesday, July 1 2009, 14-17pm
place Room C1.17, Oudemanhuispoort, Universiteit van Amsterdam (click here for directions)

speaker Stefan Frank (ILLC, Amsterdam)
title Systematic Sentence Comprehension in a World-Embedded Connectionist Model
abstract Fodor & Pylyshyn (1988) argued that connectionist models are not able to display the systematicity found in human language and thought, except by implementing a classical symbol system. This claim entails that connectionism cannot compete with the classical approach as an alternative architectural framework for human cognition. However, Fodor & Pylyshyn presupposed systematicity to result from an intrinsically systematic representational system. We present an alternative that assumes systematicity to result from systematic features of the environment, in accordance with recent views of cognition as being fundamentally embedded in the world. The viability of our idea is demonstrated by a connectionist model of sentence comprehension that does not implement a symbol system yet learns to behave systematically by relying on the structure in a microworld. The model consists in a recurrent neural network that maps sentences describing events in the microworld, onto non-symbolic representations of these events. After being trained on particular sentence-event pairs, the model can comprehend new sentences, even if these describe events on which the network was not trained. We argue that this systematicity arises robustly and in a psychologically plausible manner because it depends on structure inherent in the microworld rather than the inherent structure of the model.

speaker Franklin Chang (Hanyang University, Seoul)
title Learning to order words: A connectionist account of Japanese and English sentence production
abstract Languages can differ in striking ways. For example, the English sentence "I gave the book to him" can be translated into Japanese as "hon-o ageta" (book-ACC give). Incremental theories of sentence production have not addressed how syntax acquisition yields the representations that support adult behavior in these typologically-different languages. In this talk, I will present a connectionist model of sentence production (Chang, Dell, & Bock, 2006) that is able to learn these two languages. The model can explain the different directions of heavy NP shift in these two languages (Hawkins, 1994). The model is successful at this task, because it does not use the same universal categories for both languages (e.g., nouns, verbs, heads) and I will discuss the implications of this result for theories of syntax.

speaker Morten H. Christiansen (Cornell University, Ithaca, and Santa Fe Institute)
title A Usage-Based Approach to Recursion in Sentence Processing
abstract Most current approaches to linguistic structure suggest that language is recursive, that recursion is a fundamental property of grammar, and that independent performance constraints limit recursive abilities that would otherwise be infinite. This paper presents a usage-based perspective on recursive sentence processing, in which recursion is construed as an acquired skill, and in which limitations on the processing of recursive constructions stem from interactions between linguistic experience and intrinsic constraints on learning and processing. A connectionist model embodying this alternative theory is outlined, along with simulation results showing that the model is capable of constituent-like generalizations and that it can fit human data regarding the differential processing difficulty associated center-embeddings in German and cross- dependencies in Dutch. Novel predictions are furthermore derived from the model and corroborated by the results of four behavioral experiments, suggesting that acquired recursive abilities are intrinsically bounded not only when processing complex recursive constructions, such as center-embedding and cross-dependency, but also during processing of the simpler, right- and left-recursive structures.