Publications

Abstract

The syntactic structure of main and subordinate clauses is determined to a considerable extent by verb biases. For example, some English and Dutch ditransitive verbs have a preference for the prepositional object dative, whereas others are typically used with the double object dative. In this study, we compare the effect of these biases on structure selection in (S)VO and (S)OV dative clauses in the Corpus of Spoken Dutch (CGN). This comparison allowed us to make inferences about the size of the advance planning scope during spontaneous speaking: If the verb is an obligatory component of clause-level advance planning scope, as is claimed by the hypothesis of hierarchical incrementality, then biases should exert their influence on structure choices, regardless of early (VO) or late (OV) position of the verb in the clause. Conversely, if planning proceeds in a piecemeal fashion, strictly guided by lexical availability, as claimed by linear incrementality, then the verb and its associated biases can only influence structure choices in VO sentences. We tested these predictions by analyzing structure choices in the CGN, using mixed logit models. Our results support a combination of linear and hierarchical incrementality, showing a significant influence of verb bias on structure choices in VO, and a weaker (but still significant) effect in OV clauses

Abstract

Students who are learning to write in a foreign language, often want feedback on the grammatical quality of the sentences they produce. The usual NLP approach to this problem is based on parsing student-generated text. Here, we propose a generation-based ap- proach aiming at preventing errors ("scaffolding"). In our ICALL system, the student constructs sentences by composing syntactic trees out of lexically anchored "treelets" via a graphical drag & drop user interface. A natural-language generator computes all possible grammatically well-formed sentences entailed by the student-composed tree. It provides positive feedback if the student-composed tree belongs to the well-formed set, and negative feedback otherwise. If so requested by the student, it can substantiate the positive or negative feedback based on a comparison between the student-composed tree and its own trees (informative feedback on demand). In case of negative feedback, the system refuses to build the structure attempted by the student. Frequently occurring errors are handled in terms of "malrules." The system we describe is a prototype (implemented in JAVA and C++) which can be parameterized with respect to L1 and L2, the size of the lexicon, and the level of detail of the visually presented grammatical structures.

Abstract

A new cognitive architecture is proposed for the syntactic aspects of human sentence processing. The architecture, called Unification Space, is biologically inspired but not based on neural nets. Instead it relies on biosynthesis as a basic metaphor. We use simulated annealing as an optimization technique which searches for the best configuration of isolated syntactic segments or subtrees in the final parse tree. The gradually decaying activation of individual syntactic nodes determines the ‘global excitation level’ of the system. This parameter serves the function of ‘computational temperature’ in simulated annealing. We have built a computer implementation of the architecture which simulates well-known sentence understanding phenomena. We report successful simulations of the psycholinguistic effects of clause embedding, minimal attachment, right association and lexical ambiguity. In addition, we simulated impaired sentence understanding as observable in agrammatic patients. Since the Unification Space allows for contextual (semantic and pragmatic) influences on the syntactic tree formation process, it belongs to the class of interactive sentence processing models.