|
Internship and Thesisproposals 2007-2008
-
Modeling of Phonological Erosion
One of the main processes of linguistic change is described by
the theory of grammaticalization. One aspect of this theory
states that in certain linguistic contexts phonological
erosion/reduction can take place. This is the loss of certain
aspects of the form of a linguistic item. Some examples in
English are: "going to" -> "gonna", "let us" -> "lets",
"I will" -> "I'll". And sometimes complete words can become
grammatical markers by such processes.
The goals of this thesis would be to:
- Get acquainted with the literature on phonological
erosion/reduction during processes of grammaticalization.
- To create a computational model of these processes,
building further on models already developed at the AI-Lab.
All programming will be done in common lisp.
Relevant references:
- Hopper P.J. and Traugott E.C., Grammaticalization (Second
Edition), 2003, Cambridge UK, Cambridge University Press.
- Heine B., Claudi U. and Hunnemeyer F., Grammaticalization:
A Conceptual Framework, 1991, Chicago, Chicago University Press.
Contact: pieter [at] arti [dot] vub [dot] ac [dot] be.
-
Fluid Construction Grammar and Chart Parsing
At the AI-lab a grammar formalism called Fluid Construction
Grammar (FCG) is being developed. This formalism, although
already quite advanced, can still be improved upon
significantly and in several directions. For example, at the
moment our parsing and generation algorithms are rather
naive and greedy. As a consequence they sometimes fail to
reach complete coverage even though in principle it should
be possible. The aim of the internship and thesis would be
to incorporate more advanced techniques like chart-parsing
and chart-generation into the FCG framework.
Any student interested in this topic should be acquainted with LISP
and have a minimal interest in (computational) linguistics.
Relevant references:
-
Chapter 22 of AIMA (Russel and Norvig) on communication.
-
FCG web page
Contact: joachim [at] arti [dot] vub [dot] ac [dot] be
- Fluid Construction Grammar and Memory Based Learning
This proposal is somewhat similar to the one concerning
FCG and chart parsing. In this proposal however it would
not be chart parsing but some form of Memory-Based or
Instance-Based Learning. One of the most promising
techniques from Computational Linguistics and Machine
Learning is Memory Based Learning. The agents in the FCG
framework can recruit new capabilities when needed for
solving certain problems. It is quite easy to see that
being able to recruit Memory-Based techniques would
enhance the agents in many ways and could potentially
solve or soften problems they encounter. The main focus
of this thesis would be to investigate how these
techniques could best be combined in a multi-agent and
language-game perspective and finally to implement them
as a proof of concept in the FCG framework.
Any student interested in this topic should be acquainted with LISP
and have a interest in (computational) linguistics.
Relevant references:
Contact: pieter [at] arti [dot] vub [dot] ac [dot] be
-
Modeling of Natural Language-like Anaphora
Most natural languages exhibit constructions which allow language
users to refer back to an entity that was previously introduced
in the discourse. One simple example in English could be "Pieter
borowed me his mind." in which the possessive pronoun 'his' refers
back to entity Pieter. Many theories have been proposed how anaphora
resolution is handled and some of these have been tested in
computational linguistics. The question why anaphora are useful in
languages has received far less attention and could be solved within
the language game paradigm.
Any student interested in this topic should be willing to read
relevant literature from different domains on this topic. Basic
programming skills in Lisp are a plus as it provides the possibility
to use the frameworks we have developed at the Artificial
Intelligence Laboratory, but we are open to any other symbolic
programming language.
Relevant references:
Contact: jorisb [at] arti [dot] vub [dot] ac [dot] be
Back to Top
|
