Automatic grammar generator comprising phase chunking and morphological variation

摘要

An automated grammar generator 68 receives a speech or text segment 73. Speech segments are convened into text prior to analysis. One or more parts of the segment are identified as being suitable for natural language expression 71,72. The segment is tagged by tagger 74 to identify syntactic elements of the segment resulting in tagged headlines 75 before being parsed to parse trees 77. A phrase chunker analyses the parsed tagged headlines using variation rules (124) to determine alternative text chunks 84. Morphological analysis 92 is performed on the chunks to identify different verb tenses, such as the present paritciple, resulting in the variation of headlines 94. These variations 94 are subsequently formatted 96 before being incorporated into an automatic speech recognition, ASR, grammar model 98. Thus the present invention allows for the automatic analysis of a news item headline and determines alternative natural language expressions a user may use to reference the news item.