Precise Understanding of Language by Computers

Precise Understanding of
Language by Computers

Current Work

Ph.D. Dissertation

Title: Packed Computation of Exact Meaning Representations

Here is a draft of my dissertation defense presentation.

My dissertation is finally complete! Here it is!

Project Description

The first chapter of my dissertation.
Logic Puzzles: A New Test-Suite for Compositional Semantics and Reasoning

Some other descriptions, a bit older (and with a little overlap):

The Syntax-Semantics Interface in Glue Semantics

All the material from memos here has been integrated into my dissertation draft (above).

Here is a powerpoint presentation.

See also the Glue Semantics Bibliography.

Investigations of Linguistic Phenomena

I'm investigating certain issues in Formal Semantics, as a basis for later developing this work in Computational Semantics.

Reciprocals and their connection to "same" and "different"
- On the Syntax-Semantics Interface of "Different" and "Each Other", (with Ivan Garcia) from the 2005 Semantics Fest at Stanford.
- On the Syntax-Semantics Interface of Overt and Covert Reciprocals, a paper I wrote for a semantics seminar on quantification.
Comparatives
- Gradable Comparatives: Syntax and Syntax-Semantics Interface, a final paper I wrote for the series of two HPSG classes I took at Stanford.
- Comparative Constructions (Draft, June 18, 2005) -- extending the above paper to other kinds of comparatives and other phenomena.
Conditionals
- A Basic Analysis of Conditionals, includes an analysis of conditional sentences in the logic puzzles domain and beyond.
Events and aspect
- Some Steps in Formalizing Events, for the semantics seminar "Tense and Aspect".
Situation Semantics
- DRT and Situation Semantics, for the semantics seminar "Situation Semantics".

Application: Logic Puzzles

Logic Puzzles: A New Test-Suite for Compositional Semantics and Reasoning
Solving Logic Puzzles: From Robust Processing to Precise Semantics.
Example logic puzzles.
Examples of lexical semantic knowledge and world knowledge required for understanding logic puzzles [TBD]

Here is an old page of the project, with the NLP group.

Comparative Studies

There are many frameworks in computational semantics for scope ambiguity and underspecified representations, but not many comparative studies (although see a comparison of Hole Semantics and of MRS to CLLS here). I wrote a paper comparing Quasi-Logical Forms and Hole Semantics, and bridging them in order to benefit from their relative advantages: Decoupling Scope Resolution from Semantic Composition. I still believe that having modularity that separates semantic composition from scope resolution is the right approach, but I now think that Glue Semantics is the right tool (and not QLFs).

Presentations

Uses of Logic in Computational Understanding of English (presentation as TA in the class CS157: Computational Logic).
Syntax-Semantics Interface and Glue Semantics (presentation as TA in the class LING288/CS224U: Natural Language Understanding).