Spanish generation in the NLP system 'LOLITA'

Abstract

The aim of this research has been to modify the NLG module in the NLP system LOLITA to enable it to produce Spanish utterances. Natural Language Generation (NLG) is the production of text in a surface language by the computer in order to meet communicative goals. The NLG module of LOLITA is currently able to generate English utterances. It provides the generation capabilities required for the prototype applications built onto LOLITA. The module also aids in the development and debugging of the system as NL utterances are easier to understand than the semantic network representation. The LOLITA generator receives as in put the whole LOLITA semantic network, 'SemNet',(the system knowledge base) and adopts the traditional two components architecture. However, the distribution of task between the planner and plan-realiser (planner and realiser in other systems) differs from that in traditional systems as the plan-realiser can perform tasks such as the selection of content traditionally performed by a planner. The Spanish generator is based upon the same theoretical principles as the current English generator. SemNet forms the input of the generator and has been expanded for this purpose by the addition of Spanish lexical entries and information associated with them. The existing planner module has been used while the plan-realiser has been modified by developing new solutions where the existing ones were not adequate for producing correct Spanish utterances. The generator has been implemented in the pure functional language Haskell, taking advantage of several features of this language and, like LOLITA, it has been built following Natural Language Engineering principles. These two aspects influencing the research are also described.