Natural Language Theory and Technology

The Natural Language Theory and Technology group develops broad-coverage theories and technologies for efficiently converting between natural language expressions and machine-interpretable canonical representations. This ability is required for machines to perform such tasks as content analysis, fact-finding, question answering, inference, translation, and sense making, and to communicate naturally with people.

Our approach involves investigating the features that all human languages have in common and developing universal theoretical and computational frameworks that apply easily to languages of widely varying types. The theories of Lexical Functional Grammar (LFG) and Finite-State Morphology are examples of frameworks that we originated and that linguists and computational linguists around the world have adopted and are extending. We and our close collaborators also construct broad-coverage linguistic databases (grammars, lexicons, morphologies, semantics) that describe the distinctive characteristics of particular languages (Arabic, Chinese, English, Japanese, Turkish, Urdu, ...). Finally, we investigate and implement efficient algorithms for processing text and meaning and embed those algorithms in prototypical applications. The research disciplines in NLTT include linguistics, computer science, logic, and statistics.

Contributors

Members
Annie Zaenen
Danny Bobrow	Bob Cheslow	Cleo Condoravdi
Ji Fang	Lauri Karttunen	John Maxwell

Interns
Lucas Champollion	Liz Coppock

Administrator	Associates
Bev Ringeling	Charlotte Price	Ron Kaplan	Martin Kay

List of previous members and associates

See also the PARC pages on our natural language processing activities.

Ambiguity-enabled, Scalable Knowledge Repository (Asker)

Asker allows for natural language question answering and retrieval on massive data collections. This uses our Two-way Bridge between Language and Logic to provide a robust, broad-coverage mapping between natural language strings and abstract Knowledge Representation. The forward mapping uses the XLE parser and English ParGram grammar, followed by semantic and AKR rules using XLE's ordered rewrite system (XFR). This process is reversed for generation of grammatical, natural language strings from AKR. This research is used in part by Powerset for their indexed search products.

XLE

XLE investigates algorithms for efficient parsing and generation with broad-coverage LFG grammars, paying particular attention to the problem of ambiguity management.  These algorithms are embedded in a high-performance implementation for grammar development and natural language applications. XLE includes an ordered rewrite rule component (XFR) that is used for deeper processing (semantics, knowledge representation) and for a variety of applications.

The Parallel Grammar and Parallel Semantics project

ParGram and ParSem are part of a research and development consortium that produces large-scale LFG grammars for several languages. These grammars are developed on and are interpreted by the XLE system. They incorporate statistical machine learning techniques to induce disambiguation routines for broad-coverage constraint-based parsing.

The ParTrans (Parallel Translation) project

ParTrans develops new strategies and algorithms for translation that take advantage of the ParGram grammars, the XLE parser, generator, and rewrite system (hybridizing hand-writing and induced rules), and our general techniques for managing ambiguity.

Theoretical Linguistics

We continue our theoretical work on grammatical formalisms (especially Lexical Functional Grammar), syntactic, semantic, and knowledge representation issues, parsing and generation algorithms, and finite-state morphology and phonology. For more information, see the selected bibliography of NLTT publications.

Downloads for the NLP community

For research and education we also make available:

The Grammar Writer's Workbench for Lexical Functional Grammar, an earlier, Lisp-based implementation of LFG theory. The current XLE system is available under license for research and commercial purposes.

The PARC700 dependency bank and related tools.

Last updated: .