Multilingual Natural Language Generation (Experience from AGILE Project)1

Multilingual Natural Language Generation is an interesting and challenging field of Natural Language Processing. Automatic generation of texts in natural language could be viewed as a final part of automated translation process from one language to another. Alternative approach is given the chance with development of modern Natural Language Processing technologies, which concentrate the researchers attention on Natural Language Generation as more self-dependent area. Main idea in this respect is the claim that generation of particular text in particular language is a surface realization of particular semantic, which is communicated at the moment in the chosen language. If we have an access to (or description of) that semantic and if we use lexico-grammatical resource, which generates in another natural language we could produce a surface realization of the same semantic in the second natural language. The theoretical base for such a belief is built by functional view to the natural languages. Halliday’s functional analysis of English, Chinese and some other languages is fundamental in this respect. “Introduction to Functional Grammar” describes the basic principles of Systemic Functional Linguistics extracted by analysis of English. Beginning with the most general meta-functions of language (ideational and interpersonal) Systemic Functional Linguistics introduces the idea of communicative functions of language as common for all natural languages. One of the most successful implementations of Hallidey’s theory is the environment KPML, which inherits PENMAN system for Natural Language Generation. KPML gives the researchers basic tools for multilingual sentence generation and resource development. Multilingual generation within KPML starts from common semantics for all languages and traverses the lexico-grammatical resource of each