Unlocking topicality in text - foreground and background information in text
LLAVES 
IST-1999-12219

Technology Implementation Plan 
for LLAVES, an assessment year project under the Future and Emerging Technologies arm of
the Information Society Technologies Program

Jussi Karlgren

Summary of results

LLAVES, a project to experiment with certain types of text analysis algorithms,
met with qualified success: some of its stated aims were met, others were not,
and some were never tested to their fullest extent. LLAVES did not meet fast-track
approval for a continuation project, and the parties have decided to redirect
the results of LLAVES into other projects they are involved in, respectively.

The hypotheses of the project were that clauses in text have different
information bearing value, and that the difference can be detected from
their surface form.  This claim is well founded in theoretical work from
philology, but not uncontroversial in the field of language technology.
Our premises were that we had better base language technology than other
projects and an evaluation methodology based on large-scale information
retrieval experiments and human reader assessments. The former met with
success, the latter did not; pending better human assessment methodology
the multi-lingual evaluation was put on ice.

The methods developed specifically for LLAVES are not mature enough for
immediate deployment in commercial projects: while the LLAVES experiments
gave an indication of what type of information can be extracted from the
topical structure of clauses in texts and how it can be utilized the
algorithms are not robust enough for productification. The results merit
further study and development, and this will be done by the project
parties, most notably within the projects DUMAS and CLARITY, both funded
with the IST.

For wider dissemination, the experimental results of the project will be
published a form similar to the final report.

European interest 

The results will be strategically important in the medium-range development
of multi-lingual text understanding systems - and an area where Europe and
the EU lately has been taking an assertive role through investment in a
number of research projects. The results of the LLAVES project will be made
useful by canalization through other IST research and development projects
where they will provide valuable input: in their present form the
experiments are too specific for direct utilization in commerce or
development.