University of Twente Student Theses


Boosting LUCK: Improving the language understanding capabilities of Kaitito

Penning, M.P.J. (2007) Boosting LUCK: Improving the language understanding capabilities of Kaitito.

[img] PDF
Abstract:Kaitito is a natural language system that can be used to have conversations with one or more articial characters. All natural language systems like Kaitito have limited abilities for understanding user input. The best way to improve Kaitito's language understanding capabilities, seems to be using reformulations (paraphrases) of the user input on the level of semantic representations. This conclusion is based on the specications of the Kaitito system as well as literature from systems related to Kaitito, like chatbots and question-answer systems. The Minimal Recursion Semantics (MRS) representations used in the Kaitito system can be paraphrased using transfer rules, also called munge rules. Deriving the transfer rules by hand is quite tricky and time consuming and moreover, it does not support Kaitito's rationale of allowing the system to be authored by non-computer-linguists such as for instance English teachers. Therefore, deriving the transfer rules was automated as much as possible. Although a lot more dicult to implement than deriving the rules by hand, this automated version of deriving the transfer rules will make the system usable for non-computer-linguists. Using transfer rules involves issues like coverage and precision; rules can be too restrictive, or re on occasions they should not. The main problem with automating the transfer rule derivation process is matching two MRSs, which can result in multiple, ambiguous possible matches. Even though using paraphrases of MRSs and the semi-automatic derivation of the transfer rules have these problems inherent to them, the (preliminary) results in the domain of people introducing each other look promising.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page