a postscript version of this document
Marc N. Light
Education
Experience
| Summer 1997 - |
University of Stuttgart, Germany |
|---|
| present |
Institute for Natural
Language Processing, Research scientist -
Mats Rooth |
| |
Led the construction of a shallow but broad coverage context-free
grammar for German and trained a corresponding head-lexicalized probablistic
context-free grammar. The probablistic grammar was trained on raw
newspaper text using an EM algorithm. The primary goal of
creating the grammars was to learn a set of subcategorization frames for
verbs--the preferred frames for a particular verb can be
read off the probablistic grammar. |
| |
Continued work on hidden Markov models for
selectional preferences started in Tübingen (see below).
The selectional preferences learned by these models along with the
subcategorization frames derived from the German grammar will be
utilized by a Daimler-Benz research team as part of a speech
understanding system. |
| |
|
| 1995 - |
University of Tübingen, Germany |
| Summer 1997 |
Department of
Computational Linguistics,
Research scientist -
Steven Abney and
Erhard Hinrichs,
|
| |
Designed and implemented a system which
learns selectional preferences from shallow parsed
corpora. The system constructs a hidden Markov model for each
predicate-argument pair, in which the model's state structure is based
on a given semantic class hierarchy. These models are trained using an EM
algorithm. A training corpus consists of the word tokens that appear in
the predicate-argument pair context. From a model, one can extract
a set of semantic classes which the predicate strongly prefers. The
model can also be applied directly to tasks such as word
sense disambiguation: for each noun, the model specifies a
probability distribution over the senses of the noun. Sponsors have
accepted a project continuation proposal. |
| |
In a separate project: built a system which utilizes
partial parsing and underspecified semantic representations to
provide partial semantic analyses for spoken utterances robustly and
efficiently. The system is a pipeline consisting of an N-best
extraction system, an HMM part-of-speech tagger, a cascaded
finite-state transducer parser, and a semantic construction system.
It takes word-graphs as input and produces logical forms as
output. This system was developed as part of the
Verbmobil
project
(a speech-to-speech translation system) and was integrated into the
second Verbmobil demonstration system. Sponsors have
accepted a project continuation proposal. |
| |
|
| 1990-1995 |
University of Rochester, Rochester, NY |
| |
Department of Computer Science,
Research assistant - Lenhart Schubert |
| |
Developed an acquisition system which makes use of
fixed correspondences between derivational affixes and lexical
semantic information. The system can acquire information such as
predicate-argument structure, aspectual class, synonym and antonym
relationships between words, and verbal semantic features such as
causation and manner. This system was used to derive a lexicon from
the Brown corpus. |
| |
Developed an argument that the proper linguistic treatment of
lexical semantics phenomena requires the use of an extended version
of Standard First Order Logic. This extension should include modal
operators, predicate modification, restricted quantification, and
non-standard quantifiers. |
| |
Constructed a system which uses derivational
morphology to help deal with unknown words. The system was
integrated into one of the
Trains planning assistant project
demostration systems. The Trains project assists a user who is
solving a planning problem in a domain containing trains, cities,
factories, and products. Trains interacts with the user in spoken
English. |
| |
|
| Summer 1992 |
University of Tübingen, Germany |
| |
Department of Computational Linguistics, Research scientist -
Erhard Hinrichs |
| |
Designed and implemented an efficient algorithm for
classification in feature-based default hierarchies. Such
hierarchies are used to reduce redundancy in many NLP systems'
lexicons. The algorithm automates the task of placing entries
under the nodes in a hierarchy which minimize redundancy. |
| |
|
| 1988-1989 |
University of Zurich, Switzerland |
| |
Department of Computer Science,
Research assistant -
Marc Domenig
|
| |
Developed a formalism to represent
predicate-argument structure and its modification. This
formalism is used by
Word Manager, a
multi-purpose lexicon system designed by Marc Domenig. |
| |
|
| Summer 1987 |
BBN Laboratories, Cambridge, MA |
| |
Artificial Intelligence Laboratory, Summer intern - Varda Shaked |
| |
Developed lexical entries, semantic interpreter
rules, and knowledge base concepts for the IRUS-II natural language
front-end. IRUS-II is composed of an ATN parser, a
case-grammar-based semantic interpreter, and a terminalogical logic domain
model. |
| |
|
| Fall 1986 |
MIT Artificial Intelligence Laboratory, Cambridge, MA |
| |
Research assistant - Robert Berwick |
| |
Applied algorithms for learning k-reversible
automata to the German and English noun specifier systems. |
| |
|
| Spring & |
MIT Artificial Intelligence Laboratory, Cambridge, MA |
| Summer 1986 |
Research assistant -
Ellen Hildreth |
| |
Implemented and analyzed vision algorithms for
recovering 3-D structure of an object from a series of 2-D snapshots
of the object in motion. |
| |
|
| Summer 1985 |
National Superconducting Cyclotron Laboratory, E. Lansing, MI |
| |
Michigan State University, Research assistant - Jerry Nolan |
| |
Developed system used to design
magnets that extract beams from the cyclotron. |
Teaching Experience
| Fall 1997 |
Parsing I - Co-instructor |
|---|
| |
University of Stuttgart |
| |
Gave lectures on partial parsing, the theory of
finite state transducers, and parser evaluation. Developed
written problem sets and a programming project that required the
students to build a German named-entity recognizer and categorizer
(html).
|
| |
|
| Spring 1997 |
Introduction to Computational Linguistics - Co-instructor |
| |
University of Stuttgart |
| |
Gave introductory lectures on partial parsing with finite
state transducers, hidden Markov model part-of-speech tagging, and
probablistic context-free grammars. Developed
written and programming problem sets
(html).
|
| |
|
| Fall 1989 |
Theory of Computation - Teaching assistant |
| |
University of Rochester
(Instructor:
Lane Hemaspaandra
) |
| |
Wrote the problem set solutions and graded homework
in addition to tutoring students individually for a course on
automata, computability, and complexity theory. The course covered
most of material in Introduction to Automata Theory, Languages, and
Computation by Hopcroft & Ullman. |
| |
|
| Spring 1991 |
Natural Language Processing - Teaching assistant |
| |
University of Rochester
(Instructor:
James Allen) |
| |
Gave a third of the lectures in addition to grading
programming projects and homework for a course which introduced morphological
analysis, syntactic parsing, semantic interpretation, and discourse reasoning. |
Professional Activities
- Co-editor of the Natural Language Engineering special issue on
Lexical Semantic Tagging which is scheduled to appear in early 1999.
- Book reviewer of Corpus Processing for Lexical
Acquisition for the Journal of Logic, Linguistics, and
Information (volume 7, p. 111-114, 1998).
[Abstract,
Postscript]
- Program chair of the Workshop on
``Tagging Text with Lexical
Semantics: Why, What, and How?'' sponsored by the Association of
Computational Linguistics Special Interest Group on the Lexicon
(ACL/SIGLEX) held in conjunction with the conference on
Applied Natural Language Processing, Washington, D.C., April 4-5, 1997.
(homepage with online proceedings)
Computer Systems Experience
- Operating systems: UNIX (SunOS, Linux)
- Languages: C, Perl, Common Lisp, and Yacc/lex
Other Information
- Citizenships: USA, Swiss
- Language skills: German (fluent)
Invited Talks
- Abney, S. and Light, M., ``Hiding a Semantic Class Hierarchy
in a Markov Model,'' to be presented as part of the University of Edinburgh
Centre for Cognitive Science / Human Communication Research Centre
Seminar Programme, March, 1998.
[Abstract,
Postscript]
- Light, M., ``Truth-Conditional Semantics for Productive
Derivational Affixation: 15 Examples from English,'' presented
at the Workshop on Productive Processes in the Lexicon, which
took place in conjunction with The Annual Meeting of the German
Association for Linguistics Research (DGfS), Düsseldorf, February, 1997.
- Light, M., ``An Introduction to Machine Translation,'' presented
at the Tübingen Workshop in Computational Linguistics, Tübingen,
August, 1996.
- Light, M. and Schubert, L., ``Knowledge Representation for
Lexical Semantics: Is Standard First Order Logic Enough?,''
presented at The Future of the Dictionary Workshop, Rank
Xerox European Research Centre, Grenoble, October, 1994.
Refereed Publications and Technical Reports
- Carroll, G., Rooth, M., Prescher, D., and Light, M., ``Valence
Induction for German with a Head-Lexicalized PCFG,''
Sparkle Project Deliverable D5.1: Syntactic and Semantic Type
and Selection, 1998.
[Abstract,
Postscript]
- Light, M. and Schubert, L., ``Knowledge Representation for
Lexical Semantics: Is Standard First Order Logic Enough?'',
Proceedings of the Second International Workshop on
Computational Semantics, 1997.
[Abstract,
Postscript]
- Light, M., ``CHUMP: Partial Parsing and Underspecified
Representations,'' in Proceedings of the 12th European
Conference on Artificial Intelligence Workshop: Corpus-Oriented
Semantic Analysis, 1996.
[Abstract,
Postscript]
- Light, M., ``Morphological Cues for Lexical Semantics,'' in
Proceedings of the 34th Annual Meeting of the Association for
Computational Linguistics, 1996.
[Abstract,
Postscript]
- Traum, D., Schubert, L., Poesio, M., Martin, N.,
Light, M., Hwang, C. H., Heeman, P., Ferguson, G., and
Allen, J., ``Knowledge representation in the
TRAINS-93 conversation system,'' International Journal of
Expert Systems, special issue on Knowledge Representation and
Inference for Natural Language Processing, 9(1), 1996.
- Light, M., Reinhard, S., and Boyle-Hinrichs, M.,
``INSYST: an Automatic Inserter System for Hierarchical Lexica,'' in
Lexicon and Text: Reusable methods and resources for the
linguistic analysis of German, edited by Helmut Feldweg and
Erhard Hinrichs, Niemeyer Verlag 1996.
- Allen, J., Schubert, L., Ferguson, G., Heeman, P.,
Hwang, C. H., Kato, T., Light, M., Martin, N.,
Miller, B., Poesio, M., Traum, D., ``The TRAINS project:
a case study in defining a conversational planning agent.'' Journal
of Experimental and Theoretical AI, 7:7-48, 1995.
- Light, M., ``Classification in Feature-based Default
Inheritance Hierarchies,'' in The Proceedings of KONVENS -94,
1994.
[Abstract,
Postscript]
- Light, M., Reinhard, S., and Boyle-Hinrichs, M.,
``INSYST: an Automatic Inserter System for Hierarchical Lexica,'' in
Proceedings of the Sixth Conference of the European Chapter of
the Association for Computational Linguistics, 1993 (poster
session).
- Light, M., ``A Computational Approach to Lexical Relatedness,''
Technical Report 421, Department of Computer Science, University of
Rochester, 1993.
- Light, M., ``Rehashing Re-,'' in Proceedings of
the Eastern States Conference on Linguistics, Cornell
University Linguistics Department Working Papers, 1992.
[Abstract,
Postscript]
- Light, M., ``Taking the Paradoxes Out of Bracketing in
Morphology,'' in Proceedings of the Second Formal Linguistics
Society of Mid-America Conference, University of Wisconsin
Linguistics Department Working Papers, 1991.
[Abstract,
Postscript]
- Light, M., ``Semantic Interpretation in TRAINS-90,'' TRAINS
Technical Note 91-3, Department of Computer Science, University of
Rochester, 1991.
References
|