Antonym Generation Process Overview
The general processes for antonym generation are described below.
I. Generate aPair candidates
- APair candidates are generated from different sources (LEX, SD, PD, CC, SN).
- The new raw candidates are based from the updates of the LEXICON, Meta-Thesaurus, Lexical Tools, etc.
- The raw candidates are automatically tagged using ${ANTONYM}/antCand.data.tag.${YEAR}
- The tagged aPairs are split into:
- antCand${SRC}.data.tag
- antCand${SRC}.data.tag.tbd
=> This is the new aPair candidate file send to linguists to tag
II. Send aPair candidates to linguists for tagging
- Send antCand${SRC}.data.tag.tbd to linguist to tag
- Append the tagged file of antCand${SRC}.data.tag.tbd to antCand${SRC}.tag.tagged
III. Validate and fix tags
- Tagged file is automatically validated and fixed (for those aPairs are tagged a not canonical).
- The fixed tag file is saved at antCand${SRC}.data.tag.fixed
- Need to fix tags until tagged and fixed files are the same
- Manually copy tagged to tagged.${YEAR}
IV. Update overall tags
- Update tags from tagged.${YEAR} to antCand.data.tag.${YEAR}
- Use this updated file antCand.data.tag.${YEAR} to re-run all steps until there is no change.
V. Generate Release
Generate release files from antCand.data.tag.${YEAR}
- antonyms
- negation cue words
VI. Status on Antonym Source Models
Source | Descriptions | Status
|
LEX | from lexRecords with negative code | annaul Lexicon updates
|
SD | from suffixD with negative tags | annaul Lexicon updates
|
PD | from prefixD with negative tags | yet to complete previous Lexicon releases
|
CC | from collocates in Corpora (MEDLINE) | annaul MEDLINE updates
|
SN | from semantic network (WordNet) | yet to complete WordNet 3.0
|