Boosting a Chatterbot Understanding with a Weighted Filtered-Popping Network Parser

Javier M. Sastre Martinez (1) , Jorge Sastre (1) , Javier Garcia-Puga (1)

1 LIGM - Laboratoire d'Informatique Gaspard-Monge

We describe here an application of the filtered-popping network (FPN) parser in (Sastre, 2009a) for boosting the recognition capabilities of an AIML (Wallace, 2004) chatterbot: the MovistarBot. This conversational agent was developed by Telef'onica R&D as an attractive medium for the request of mobile services, such as sending SMSs or downloading games, accessible via short text messages in Spanish through MSN Messenger. AIML being too cumbersome for the fine description of complex sentences, the original chatterbot required services to be requested following a strict command syntax; natural language (NL) requests were answered with the description of the corresponding command syntax, assumed by the presence of keywords. We have manually constructed a recursive transition network (RTN) with output recognizing and tagging a significant variety of requests in Spanish, implemented an automatic RTN weighting procedure for ambiguity resolution and adapted the FPNparser for the automatic translation of the RTNsentences into command requests.

Domaines: Informatique et langage [cs.CL]

Javier M. Sastre Martinez, Jorge Sastre, Javier Garcia-Puga (Dir.). Boosting a Chatterbot Understanding with a Weighted Filtered-Popping Network Parser. 4th Language & Technology Conference (LTC'09), Wydawnictwo Poznańskie Sp. z o.o., pp.74-78, 2009. ⟨hal-00636996⟩