Analyzing telegraphic messages

Abstract

We have shown how highly telegraphic messages can be analyzed through straightforward extensions of the mechanisms employed for the syntactic and semantic analysis of standard English text. We have extended previous work on the grammatical analysis of telegraphic messages by allowing for the omission of function words as well as major sentence constituents. This substantially increases syntactic ambiguity, but we have found that this ambiguity can be controlled by applying semantic constraints during parsing and by using a "best-first" parser in which lower scores are associated with analyses which assume omitted function words. To recover missing arguments from telegraphic text, we have adopted a strategy in which such omitted arguments are treated as anaphoric elements. In order to resolve anaphoric ambiguities, we have extended the anaphora resolution procedure to take account of the implicit causal and enablement relations in the text. We generate alternative resolutions of anaphoric reference and then select the text analysis with the highest "coherence": the analysis for which we can identify the greater number of intersentential relations.