Project in Natural Language Generation

DAVE: open the pod bay doors, HAL.

HAL: I am sorry, I can not do that.

How did HAL say that?

The reason behind HAL's response is part the whole story line of 2001 An Space Odyssey. However we may expect that before saying that HAL's may have built something like:

$\begin{displaymath}\left[ \begin{array}{ll} cat & clause \\ epistemic-modali... ...ray} \right] \\ \end{array} \right] \\ \end{array} \right] \end{displaymath}$

These kind of constructions are the input of the FUF software packaged, developed at Columbia University by Michael Elhadad during his PhD thesis.

The system has been widely deployed. The first implementation was done in Common Lisp. The lastest version (CFUF) is a C implementation embedded in an Scheme interpreter. It targets one of the drawbacks of the initial version (namely, execution speed). On the other hand, the data structure and programming interface is still tied to Lisp.

The goal of this project is to improve the interface to FUF in the following directions:

Provide an embedding of the existing CFUF into the Java Virtual Machine by means of the Java Native Interface (JNI). The definition of an Application Programming Interface (API) suitable for working with FUF would be a first step in the whole migration of the system to a whole re-implementation in Java (a putative JFUF).
Provide a better encoding for the input of the FUF system, both from the perspective of language independence and/or supporting tools. An XML encoding of structures like the ones shown above would be of interest.

This project is a very nice opportunity to get involved with the not-so-well-known field of Natural Language Generation, while getting hands-on experience with real-life JNI programming. There is plenty of space in this project for the student to develop personal skills and creativity, for instance, on the design of the target API.

We have our first deliverable, an alpha release of an API to access FUF/SURGE functionality in Java, written by Howard Chu (hhc42@columbia). It is available for download. Take a look at known issues before starting to work with it. You can also browse the JavaDoc documentation online. A mailing list is also available.

Please contact Pablo Duboue (pablo@cs) for further information.

Last update: Fri Sep 20 06:53:30 EDT 2002