|
|
Slides &
Reports of Bachelors in Computational Linguistics |
|
|
| X--Bar theory: A general scheme for
Phrase Structure Rules Seminar (Summer Semester 2003) on Formal Syntax in Chomsky Tradition. This seminar dealt with formal syntax in Chomskian tradition. We concentrated on the X--bar theory which is one of the fundamental components of the Government and Binding theory as proposed by Noam Chomsky. We explain the theory formally, as well as the motivation behind the X--bar theory by considering the common patterns in the phrase structures and sentences. We also discuss the extension of X--bar theory to different languages. The report. |
| Collaborative Problem Solving in Dialogue
Models for Practical Purposes: Seminar (Winter Semester 2003-2004) on Modelling Natural Dialogue for Practical Applications. Here, we study and analyze existing dialogue models that are designed for practical purposes. The dialogue modeling issues of collaborative problem solving are discussed in detail in this seminar. We analyze the architecture of Collagen, a middle ware agent based on the collaborative discourse theory. We also discuss specific agents based on Collagen like PACO (intelligent tutoring agent). Collaborative agents based on multi--tasking is another aspect of this study. A specific intelligent agent WITAS , an unmanned vehicle, based on concept of multi--tasking is also discussed. The report. |
| Off--line measures in Human Speech Production:
Seminar (Summer Semester 2004) on Human Speech Production. This seminar falls under Psycholinguistic study of Human Speech Production processes and human cognition. We discuss three papers on off--line/unconscious processes of speech production. We see arguments for and against the independence of the stages of the grammar as well as the phonological production modules. The experiments took samples from the everyday speech errors like slip-of-the-tongues where people mispronounces some words resulting in the formation of nonsensical or legitimate words and errors like tip-of-the-tongue states where the speaker reports that he/she knows the word but cannot retrieve the phonological form of the word to articulate. The report. |
| Speaking Robots: Black Jack Dealer -
A Robot that speaks and plays cards. Software Project (Summer Semester 2004) on Speaking Robots. In this project, our team (of three) built a robot that can play the card game of Black Jack against a user / player as well as engage in a dialogue (in German) specific to the domain of this game. We designed and developed within the duration of 7 weeks, a basic version of the Black Jack dealer to play with one player and then extended with one more module which can sort the cards at the delivery point. We also discuss some future extensions for the robot, so as to accomadate more card games. The body of the Robot was made using Robotic Invention System kit of LEGO mindstorms building blocks. The Robot was then equipped with a small dialogue system which was developed as a commercial software called Diamant Dialogue System developed at the CLT Sprachtechnologie GmbH, Germany. Though the basic kit provides a graphical programming system for the central processing unit (RCX), we used a flexible third party RCX programming called LeJOS which allows a fine tuned programming using JAVA. In this project, we tried to incorporate the image recognition software as well as a dialogue system that enables communication between the robot and the human player. It is possible to extend the basic model of card--robot so that it can deal with more card games like Poker, Bridge etc. The project report. |
| Bachelor's Thesis: Evaluation of the
stochastic extension of a constrain-based dependency parser. Bachelor's Thesis (Summer Semester 2004) on Evaluation of the stochastic extension of a constrain-based dependency parser . This thesis presents the results of an experimental evaluation of a stochastic extension of the Extensible Dependency Grammar (XDG) solver, a constraint--based dependency parser. XDG is a formal framework for dependency grammar which supports the characterization of linguistic information along multiple dimensions of description for phenomena like linear precedence, predicate argument structure, quantifier scope structure etc. This parser uses A* search strategy to prune large search space of partial syntax trees and gives the best first parse within the least possible time. We tested the performance of the parser on sentences and grammars automatically acquired from the Penn Treebank and evaluated how stochastic guidance (acquired from the frequency informations from the Penn Treebank corpus) helps the parser to prune the search tree. As the performance parameters for the performance evaluation, we used: time taken for parsing, the maximum size of the agenda maintained by the A* search regime, the number of failed nodes in the search tree and number of solutions encountered during the traversal of the search space. The stochastically enhanced XDG (SXDG) parser's search strategy is significantly superior compared to the exhaustive search approach, which enumerates all solutions. We believe that techniques like smoothing (or statistical discounting) and also modifying the constraint solver such that constraint distribution is guided by the statistical information from the corpus could further improve the performance of the parser. The thesis. |