According to the definition provided by Systemic Functional Linguistics, the texture of a text is related to the listener`s perception of coherence and is manifested by a set of semantic relations, called cohesive ties, holding between text chunks. Coreference is one of the most studied ties, but many other relations deserve attention. This paper presents a module, called `Texture Resolution Module` (TRM), which attempts to identify the relevant anaphoric semantic relations linking the current sentence to the preceding ones. TRM tracks the entities mentioned as long as they are introduced in the discourse and uses a set of declarative rules to guess which ties hold for a certain referring expression. The architecture designed for TRM highly emphasizes system modularity and resource reuse: new rules can easily be added to deal with new linguistic phenomena encountered in the domain, allowing for an incremental tuning of the module. Rules can be written independently to one another, assigning to each of them a confidence score that expresses the certainty of the guess made by the rule. Some of the rules have general validity and can be applied across different domains
The Texture Resolution Module: a General-Purpose Customizable Anaphora Resolutor
Not, Elena;Zancanaro, Massimo
1998-01-01
Abstract
According to the definition provided by Systemic Functional Linguistics, the texture of a text is related to the listener`s perception of coherence and is manifested by a set of semantic relations, called cohesive ties, holding between text chunks. Coreference is one of the most studied ties, but many other relations deserve attention. This paper presents a module, called `Texture Resolution Module` (TRM), which attempts to identify the relevant anaphoric semantic relations linking the current sentence to the preceding ones. TRM tracks the entities mentioned as long as they are introduced in the discourse and uses a set of declarative rules to guess which ties hold for a certain referring expression. The architecture designed for TRM highly emphasizes system modularity and resource reuse: new rules can easily be added to deal with new linguistic phenomena encountered in the domain, allowing for an incremental tuning of the module. Rules can be written independently to one another, assigning to each of them a confidence score that expresses the certainty of the guess made by the rule. Some of the rules have general validity and can be applied across different domainsI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.