Solving crosswords using the Web
We designed and implemented a software system, called WebCrow,
that represents the first solver for Italian crosswords
(note: it can be easily extended to other languages) and
the first system that tackles a language game using the Web as knowledge base.
WebCrow does not posses any
knowledge-specific expert module, but only a limited set of useful
modules, which includes a dictionary and a small database (one order of
magnitude smaller than Proverb's).
It has to be noticed that the system described in AAAI-05 is basic and
we are currently working on it. Nevertheless the performances of WebCrow
have already been interesting, as they are comparable to an average
human player (a recent experimental test has showed that it outperforms
the level of undergraduate students). WebCrow's future challenge is to
be competitive against human masters.
WebCrow is mostly language-independent and it can aspire to be able of
solving crosswords in different languages (currently, it has been tested
on Italian crosswords). Furthermore, the Web provides the most rich and
self-updating repository of human-knowledge that can help to enfold with
semantics real-life concepts.
With this approach we can avoid to implement various different knowledge
databases (eg. geography, cinema, history, etc.), which can require a
great developement effort.
We are confident that the technologies developed for cracking crosswords
will be useful also for solving real-world problems, especially those
involving the process of Web information.
Leaving aside a justified cognitive curiosity towards the problem, the
development of a technology able to face this challenging task provides
significant insights in real-world applications like automatic
dictionary creation, support for text-summarization (finding the exact
word that answers a definition could represent a possible approach to
paraphrase compression), prompting aid writing (in text editors),
support for standard question answering and information extraction.