What is VINCI?

VINCI is a natural language generation environment which has been evolving since 1986.

Originally conceived to create drill exercises for language learning, the system provides linguists with a collection of linguist-friendly metalanguages allowing them to model natural language phenomena and to generate utterances based on their models. Aspects of the metalanguages deal with semantics, syntax and morphology (both inflectional and derivational), as well as lexical relations.

The two principal tasks of the VINCI system are sentence generation and word creation. Sentence generation involves the creation of phrases or sentences in a language specified by the user. In simple cases, these sentences are created at random. In our current research, however, we are experimenting with the use of "semantic expressions" to control their content.

Word creation involves the systematic application of word-formation rules (lexical transformations) to an existing lexicon to obtain all possible new forms which the rules specify.

To date, VINCI has been used to examine a broad range of linguistic phenomena including pronominalization, coordination, suffixation, linguistic humour, and second language performance errors. (See list of publications.) It has been used in French, English, Italian and Russian, though predominantly the first.

In the realm of linguistic modelling, for example, the linguist writes a more or less complex grammar for some part of a language and generates sample utterances to evaluate the model. In this way, VINCI provides a third alternative for empirical work, along with corpus studies and linguistic intuition. The same applies to experiments in word-formation. In this last area, VINCI has also been used to compare the productivity of different rules by generating random instances and asking native speakers to evaluate the products.

At a more complex level, VINCI can generate clusters of utterances, allowing it to function as a performance testing environment. If we wish to test prepositional usage among L2 learners of French, for example, we can cause the system to generate questions which will elicit the use of certain prepositions in their responses. The system can also quietly generate one or more expected responses for comparison with learner input. The same features also allow VINCI to be used as a generative CALL environment.

A comparison algorithm has recently been developed which provides a detailed list of differences between a "correct" sentence and student response, including changes of word order as well as lexical, morphological, phonological and typing errors. We are currently working on adaptive testing processes to lead us from these symptoms to a diagnosis of a learner's underlying misconceptions.

The VINCI facilities are embedded within ivi, a powerful editor capable of manipulating text and records (and indeed, in some versions, a variety of other objects).

As indicated elsewhere in these pages, we are happy to offer this software to other researchers who want to make use of it.