This thesis describes a computational model for interpreting natural language expressions in an interactive multimodal query system integrating both natural language text and graphic displays. The primary concern of the model is to interpret expressions that might involve graphical attributes and expressions whose referents could be objects on the screen. Graphical objects on the screen are used to visualise entities in the application domain and their attributes (in short, domain entities and domain attributes). This is why graphical objects are treated as descriptions of those domain entities/attributes in the literature. However, graphical objects and their attributes are visible during the interaction, and are thus known by the participants of the interaction. Therefore, they themselves should be part of the mutual knowledge of the interaction. This poses some interesting problems in language processing. As part of the mutual knowledge graphical attributes could be used in expressions, and graphical objects could be referred to by expressions. In consequences, there could be ambiguities about whether an attribute in an expression belongs to a graphical object or to a domain entity. There could also be ambiguities about the referent of an expression is a graphical object or a domain entity. The main contributions of this thesis consist of analysing the above ambiguities, designing, implementing and testing a computational model and a demonstrational system for resolving these ambiguities. Firstly, a structure and corresponding terminology are set up, so these ambiguities can be clarified as ambiguities derived from referring to different databases, the screen or the application domain (in short, source ambiguities). Secondly, a meaning representation language is designed which explicitly represents the information about which database an attribute/entity comes from. Several linguistic regularities inside and among referring expressions are described so that they can be used as heuristics in the ambiguity resolution. Thirdly, a computational model based on constraint satisfaction is constructed to resolve simultaneously some reference ambiguities and source ambiguities.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:652279 |
Date | January 2001 |
Creators | He, D. |
Publisher | University of Edinburgh |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Page generated in 0.002 seconds