Information about the sense tagged corpus
Hwee Tou Ng
Dec 1996

This directory contains the sense tagged corpus hand tagged by 12
undergraduates from the Linguistics Program of the National University
of Singapore. The corpus contains sentences in which about 192,800
word occurrences have been tagged with WordNet senses. These sentences
are taken from the Brown corpus and the Wall Street Journal corpus,
and the sense-tagged word occurrences consist of 121 nouns and 70
verbs which are the most frequently occurring and ambiguous words of
English. These 121 nouns and 70 verbs are listed in nlist.txt and
vlist.txt respectively. The WordNet 1.5 sense definitions of these
nouns and verbs that were used in preparing the sense-tagged corpus
appear in ndefs.txt and vdefs.txt respectively.

All sentences that contain occurrences of a particular "word" in part
of speech noun (or verb) are collected together in one file "word.n"
(or "word.v"). Sentences in these files are arranged into one sentence
per line, separated by a blank line. We illustrate the format of a
sentence with the following example taken from the first line of the
file "action.n":

ca01.db #020 `` These >> actions 8 << should serve to protect in fact and in effect the court 's wards from undue costs and its appointed and elected servants from unmeritorious criticisms '' , the jury said .

Each sentence starts with a file identification (ca01.db in this
example) and a sentence number (#020 in this example), followed by the
actual sentence. The word that is sense tagged is delimited by ">>"
and "<<", and the tagged WordNet sense number appears after the word
before the closing "<<". So in the example sentence shown here, the
word occurrence "actions" is tagged with sense 8 of the noun "action"
in WordNet. This sentence is taken from file "ca01.db" of the Brown
corpus, and it is the 20th sentence in this file. The file
identification of a Brown corpus file starts with "c???.db" while that
for a Wall Street Journal file starts with "dj??.db".

Sometimes a verb is followed by a particle such that none of the
WordNet sense definitions for the verb apply. The verb together with
the particle is more appropriately treated as a collocation. For
example, in the sentence:

"He put off the meeting to next month."

"put off" (as in delay) is more appropriately treated as a
collocation, and none of the given WordNet senses of "put" apply
here. In such a situation, sense -1 is assigned to the verb occurrence
"put".

A word is assigned sense 0 if (1) none of the given WordNet senses is
appropriate and it is not part of a verb collocation; or (2) it is not
possible to assign it a unique sense based on the context of the given
sentence; or (3) there is a genuine mistake in the sentence (eg., the
word to be tagged appears in the wrong part of speech).

This sense tagged corpus was first reported in the following paper at
ACL-96:

Integrating Multiple Knowledge Sources to Disambiguate Word Sense: 
An Exemplar-Based Approach
Hwee Tou Ng and Hian Beng Lee
In Proceedings of the 34th Annual Meeting of the Association for
Computational Linguistics, pages 40-47, Santa Cruz, California, USA,
June 1996.
http://xxx.lanl.gov/abs/cmp-lg/9606032

For further questions about the corpus, please contact:

Dr. Hwee Tou Ng
Defence Science Organisation
20 Science Park Drive
Singapore 118230
Republic of Singapore
email: nhweetou@dso.gov.sg