LLAVES

Unlocking topicality in text - foreground and background information in written language

One-year assessment project funded by the Information Society Technologies Program under its Future and Emerging Technologies Open scheme.

LLAVES investigates characteristics of clauses in written text with the objective of distinguishing different types of clause.

Clauses: Clauses have different information bearing roles.
Mechanisms: The role of a clause is conveyed from author to reader by their surface form or position.
Generality: Clausal roles are not language specific, but heavily text ecology and domain specific.
How? Different sorts of clause can be coaxed apart using language and domain specific mechanisms.
Use: These mechanisms can be applied in the development of practical tools.

The overall success criterion of the project is to find that any or all of the following apply:

there are systematic syntactic differences between foreground and background clauses as determined by kappa and Mann Whitney U tests;
indexing material separately for foreground and background clauses gives better results than state-of-the-art indexing as determined by TREC evaluation;
there are systematic likenesses between foreground and background clauses across languages in translated texts according to the kappa test; and
summarizing texts by selecting foreground vs. background clauses gives a basis for multi-doc summaries, given that human assessors give quality judgments of summaries, again as determined by the kappa test.

All items are quantitative; the first two items have built-in thresholds for hypothesis testing.

During the course of the project we found good reason to concentrate on retrieval evaluation rather than evaluating summarization competence.