Lexicon in Text Generation

摘要

This report compares several lexicons used in computational text generation systems, with respects to the size of the lexical item, the way occurrence phenomena are represented, and the way semantic information is included. The lexicons examined can be roughly divided into two principal groups with respect to the size of the item, phrasal lexicons and word-based lexicons. Phrasal lexicons, which are more numerous, have large units (sometimes whole sentences) stored as lexical entries. They often tend to represent syntactic structure within the lexical item, and may also contain variables or slots which can be filled by other items. This type of lexicon generally provides the primary line between semantic and syntactic representation by mapping semantic structures onto syntactic structures. The word-based lexicon, on the otherhand, merely inserts words into previously built syntactic structures, using feature specifications to guide the process. Lexicons also vary with respect to the amount of occurrence information they contain. Most lexicons represent subcate; organizational (argument structure) information, either by means of features or with syntactically labelled slots. They can also have noncompositional multi-word units (idioms) as lexical entries. Some lexicons represent selectional information as well, by means of semantic feature restriction on slots. collocational information is rarely included. The meaning of a lexical item can be indicated by a pointer to a concept in a semantic network or by a pattern which matches a piece of conceptual structure. Some systems additionally have a concept of lexical choice, i.e., routines which explicitly choose between alternative lexical realizations of a particular meaning.