An example based approach for parsing natural language sentences

Development of a parser from scratch is typically time-consuming and error prone. A more efficient approach may lie in adapting an example-based parsing scheme, provided an appropriate strategy is developed for knowledge elicitation from the example base. This paper proposes such a scheme for developing a Link Grammar based parser The proposed scheme extracts knowledge from an already parsed example base in the form of "link information" and "phrase templates". For a given input sentence the parsing algorithm first extracts the possible links for the constituent words from the link dictionary. Then using the phrase templates and the possible links of the words, the sentence is parsed bottom-up. Algorithms have been developed to identify different phrases in the input sentence and to handle unknown words. The examples considered in this work are for English sentence. The proposed scheme uses very little language-specific information. Hence, the proposed algorithms can be adapted for other languages as well.