(353) previous ~ index ~ next

To: Thorsten Brants <brants@parc.xerox.com>
From: Jonathan Fiscus <jonathan.fiscus@nist.gov>
Subject: Cross-lingual NED
Date: Tue, 11 Jun 2002 08:13:58 -0400

Thorsten,

Thorsten Brants wrote:

> I have a question concerning the languages in TDT-2002:
> the call mentions English, Chinese, and Arabic, but it did not become
> clear to me whether all five tasks will be done for all three languages,
> and whether there are any cross-lingual tasks. I am especially interested
> in the novelty task. Will this be for all three languages or for English only?

Currently, only the New Event Detection task is monolingual. The
evaluation plan is the authoritative resource. We limited the language
domain for NED to English for a couple reasons:

1. The task is very difficult and there's sufficient room for
improvement in the monolingual setting.

2. The number of target new events, one per annotated topic, is small a
set of trails. If we were to do a multilingual version, the set of
topics with English first stories would be reduced and the topics with
Mandarin or Arabic would be even smaller.

Do you still think cross-lingual NED would be a valid task? Remember
we'll only have 60 new topics in the TDT4 collection. If I were to
speculate about the likelihood of a non-English first story, it would be
very small.

Jon
-------------------------------------------------------------
To unsubscribe from tdt-distrib, email majordomo@ldc.upenn.edu
with "unsubscribe tdt-distrib" in the body of the message.
(353) previous ~ index ~ next

Last updated Wed Jun 19 11:58:04 2002