The Mikrokosmos Semantic Analyzer

Stephen Beale's Homepage
Kavi Mahesh's Homepage
Mikrokosmos Homepage
Computing Research Lab Homepage
The Mikrokosmos Machine Translation System is controlled by an architecture called HUNTER-GATHERER (HG). HG was developed at NMSU in response to the ever-increasing demands being placed on natural language systems due to: 1) increased lexicon coverage (including one-to-many mappings between words and word senses), 2) unrestricted text inputs, and 3) increased semantic coverage of textual features such as discourse, ellipsis, coreference, metonymy and metaphor.

HG integrates three related AI search techniques and applies the result to processing computational semantics, both in the analysis of source text to discover underlying semantics, as well as in the planning of target text using input semantics. The approach is summarized in the name "HUNTER-GATHERER":

--- Branch-and-Bound and Constraint Satisfaction allow us to "hunt down" non-optimal and impossible solutions and prune them from the search space.

--- Solution Synthesis methods then "gather together" all optimal solutions while avoiding exponential complexity.

Below we present tutorials on HG, as well as on the basic methodology of knowledge-based semantic disambiguation. Following the tutorials are several papers written by the analyzer team (Stephen Beale, Kavi Mahesh and Sergei Nirenburg). Here is a summary of the results of analyzing several large Spanish texts.

TUTORIALS

TUTORIAL 1: HUNTER-GATHERER or Download ps file

TUTORIAL 2: BASIC KNOWLEDGE-BASED SEMANTIC DISAMBIGUATION

PAPERS

Here are some papers that describe the Mikrokosmos Semantic Analyzer:

Basic processes: setting up constraints, disambiguating, etc.. This paper is a little out of date with respect to the control methodology,but it is a good introduction to how knowledge-based semantic disambiguation works:

Beale, Stephen, Sergei Nirenburg and Kavi Mahesh. 1995. Semantic Analysis in the Mikrokosmos Machine Translation Project. In Proc. Symposium on Natural Language Processing. Kaset Sart University. Bangkok, Thailand.
download ps file

Control: how we avoid exponential complexity.

Beale, Stephen. 1997. Hunter-Gatherer: Applying Constraint Satisfaction, Branch-and-Bound and Solution Synthesis to Computational Semantics. Ph.D. Dissertation, Language Technologies Institute, School of Computer Science, Carnegie Mellon University. 1997.
download ps file

Beale, Stephen. 1997. Using Branch-and-Bound with Constraint Satisfaction in Optimization Problems In Proc. AAAI-97. Providence, Rhode Island.
download ps file
This paper copyright AAAI

Beale, Stephen, Sergei Nirenburg and Kavi Mahesh. 1996. HUNTER-GATHERER: Three Search Techniques Integrated for Natural Language Semantics. In Proc. AAAI-96. Aug. 4-8, 1996. Portland, Or.
download ps file
This paper copyright AAAI

Beale, Stephen. 1996. HUNTER-GATHERER: Applying Constraint Satisfaction, Branch-and-Bound and Solution Synthesis to Computational Semantics. Technical Report MCCS-96-289, Computing Research Lab, New Mexico State University, Las Cruces, New Mexico.
download ps file

Text Planning: we are just beginning work on text planning. Here are a few papers, mostly related to control:

Beale, Stephen and Sergei Nirenburg. PICARD: The Next Generator. In Proc. 8th International Workshop on Natural Language Generation, Poster Session. June 13-15, 1996. Sussex, UK.
download ps file

Beale, Stephen and Evelyne Viegas. Intelligent Planning Meets Intelligent Planners. In Proc. ECAI-96, Budapest. 1996.
download ps file

Beale, Stephen and Sergei Nirenburg. 1995. Dependency-Directed Text Planning. In Proc. IJCAI-95 Workshop on Multilingual Text Generation. Montreal, Canada.
download ps file

Graph Topology: here is a draft paper (please do not quote yet) relating the HUNTER-GATHERER control methodology to general computer science topics. In particular, I examine how it can be used in graph coloring problems (which can be related to various real-world problems such as scheduling). I examine the relationship between a problem's topology and the effectiveness of HUNTER-GATHERER:

Beale, Stephen. 1996 (DRAFT FOR SUBMISSION). Exploiting Graph Topology for Optimization Problems. Submitted to Constraints Journal.
download ps file

Mikrokosmos Publications. Publications from the Mikroksomos team covering the lexicon, ontology, analyzer, as well as theoretical background of our knowledge-based approach to machine translation.

JAVA DEMO

Here is the beginnings of a java web DEMO.

Copyright © 1997, Stephen Beale.

Last revised March, 1997. Send questions and comments to Stephen Beale.