• One line of inquiry seeks to establish the nature and extent of the constraints on sequential learning and processing of hierarchical structure. This research centers on using artificial languages to study the limitations on human sequential learning and processing, addressing issues related both to implicit learning and language acquisition. 

• One project is exploring how constraints on sequential learning can help explain the existence of language universals. Christiansen & Devlin (1997) presented converging evidence from a theoretical analysis of rule interactions, connectionist simulations, and typological language data supporting a functional account of basic word order universals (i.e., branching direction) couched in terms of nonlinguistic constraints on learning and processing. Results from an artificial language learning experiment, which tested predictions from the connectionist simulations, have provided further support for this account (Christiansen, in preparation-a). More recently, additional experiments have successfully extended this line of research to the more complex sequential structure exemplified by subjacency in natural language (Ellefson & Christiansen, 2000b). An overview of this work is found in Christiansen & Ellefson (2002). Future work will seek to develop a new methodology for studying artificial language learning using ERP, promising to provide a more sensitive measure of sequential learning abilities in both adults and infants.
• A another project (in part motivated by the ideas outlined in Morrison & Christiansen, 1995) tested agrammatic aphasic patients and normal controls on an artificial language learning task. The main goal of this investigation is to gain insight into the kind of statistical information that agrammatics may be able to learn and process (Christiansen, Kelly, Shillcock & Greenfield, in preparation). The results indicate that agrammatics not only have problems with language, but also with sequential learning.

• A second line of research investigates the degree to which such nonlinguistic constraints can help explain performance limitations in human language processing. This work explores the role of sequential learning in the processing of different types of sentence structure, using both connectionist modeling and psycholinguistic experimentation.

• In one project, a self-paced reading task (with on-line grammaticality judgments) was used to test novel predictions emerging from a connectionist model (Christiansen, in preparation a) concerning the processing of sentences with multiple instances of the same recursive structure (Christiansen & MacDonald, in preparation). The results suggest that there are intrinsic constraints not only on center-embedded recursion, but also on the simpler left- and right-branching types of recursion. This perspective is pursued further in a recent paper that describes the potential impact of connectionist language processing on the notions of constituency and recursion in the psychology of language (Christiansen & Chater, in press). Related work on recursive sentence processing in connectionist networks was reported in Christiansen & Chater (1999).
• This framework has also been extended to explain individual differences in sentence comprehension in terms of a combination of experiential and biological factors, offering a reappraisal of the role of a fixed working memory capacity in accounting for individual differences in sentence comprehensionA precursor to this work is found in Christiansen & MacDonald (1999), while the theoretical groundwork is provided in MacDonald & Christiansen (2002). Predictions from this new approach is currently being investigated via a comprehensive training study.

• A third line of inquiry links the first two lines of research directly to language acquisition, aiming to determine the kind of sequential learning device required for the acquisition of language. This research focuses on how the integration of multiple sources of probabilistic information by a sequential learning device may facilitate language acquisition. 

• One project is looking at the relationship between speech segmentation and sequential learning. Past work have demonstrated that recurrent connectionist networks provide a good model of early infant word segmentation (Allen & Christiansen, 1996; Christiansen, 1998; Christiansen, Allen & Seidenberg, 1998) - even under noisy circumstances (Christiansen & Allen, 1997). This work has recently been extended to cover issues related to statistical learning (Christiansen & Curtin, 1999a, b). Predictions derived from this work have been corroborated by statistical learning experiments using adult subjects (Christiansen, Conway & Curtin, 2000, in preparation). Other experiments are currently being designed to test additional predictions concerning the integration of multiple probabilistic information sources in speech segmentation. A comprehensive review of the above segmentation work can be found in Christiansen & Curtin (submitted).
• We have just begun a new project that seeks to extend the multiple-cue approach to the learning of grammatical categories and basic aspects of syntax. Teams from the US, UK, France, and Japan will investigate the role of integrating subtle statistical cues in language acquisition. Corpus analyses, psycholinguistic experimentation, computational modeling, and ERP studies are planned. Currently, we have established the computational feasibility of the multiple-cue approach (Christiansen & Dale, 2001).  See the Multiple-Cue Integration in Language Acquisition site for more details.

• The above three lines of research are inspired by an overarching perspective on the evolution and acquisition of language. 

• A comprehensive theory of the acquisition, processing and evolution of language is currently in the process of being fleshed out in the form of a book entitled Creating language: Towards a unified framework for language acquisition, processing and evolution (Christiansen & Chater, in preparation-a). This book brings together and interpret important recent findings from evolutionary approaches to language, behavioral and computational studies of language acquisition and processing, providing a unified perspective of the nature, evolution and acquisition of language.
• Within this overall perspective, some of the lab research focuses on the role of sequential learning in the evolution of language. This research suggests that the evolution of language may have been constrained by sequential learning and processing mechanisms existing prior to the emergence of language (Christiansen & Chater, in preparation-b; Christiansen, Dale, Ellefson & Conway, 2002). This suggestion is supported by a review of the sequential learning abilities in non-human primates (Conway & Christiansen, 2001).
• Our research on language evolution also includes a an edited volume in progress, Language evolution: The states of the art, comprising a collection of original "position papers" by Fiveteen major contributors to the rapidly growing field of language evolution (Christiansen & Kirby, in preparation).

• Connectionist modeling forms a crucial and highly integrated part of the research in the lab. Part of our research therefore centers around a more general appraisal of the connectionist approach to language and cognition.

• This work has lead to the completion of a Special Issue of Cognitive Science on Connectionist models of human language processing: Progress and prospects (Christiansen, Chater & Seidenberg, 1999), assessing the progress made so far and the prospects for future development of connectionist models of natural language processing. Click here to see a preface and table of contents and here to see abstracts of the papers. An extended version of the Special Issue is being finalized as a separate book entitled Connectionist psycholinguistics (Christiansen & Chater, 2001e) to be published in 2001 by Ablex Publishers. A comprehensive introduction and review of connectionist modeling of language processing can be found in Christiansen & Chater (1999a, 2001b) and Chater & Christiansen (1999).
• A new goal for connectionist language modeling has been put forward under the banner "connectionist psycholinguistics", challenging connectionists to provide closer fit with psycholinguistic data (Christiansen & Chater, 2001a). An appraisal of connectionist work on speech processing within this framework is found in Chater & Christiansen (in press). Similar considerations also apply to other aspects cognitive modeling, as exemplified by a critical review of connectionist models of developmental cognitive disorders (Conway, Schaper, Ellefson & Christiansen, in preparation).

Please email suggestions/errors to mhc27@cornell.edu