DeConf

De-conflated Semantic Representations

Sense representations generated using the algorithm introduced in:

• M. T. Pilehvar and N. Collier, De-Conflated Semantic Representations. EMNLP 2016, Austin, TX.

Download

Semantic representations for around 130K word senses in WordNet 3.0:

• Download bin: in bin format (240 MB), or
• Download txt: as a compressed (bzip2) text file (345 MB).

The sense representations live in the same 300-dimensional semantic space of words and phrases trained by Word2vec on the Google News Corpus (direct link).

Alternatively, you can download the 50-dimensional vectors that live in the Wikipedia 2014 + Gigaword 5 space of GloVe:

• Download txt.bz2: in txt format (56 MB), or

Data format

sense_ID    <a real valued vector>

where sense_IDs are in the form word#1, word#2, ..., word#n (where n is the polysemy of the word).

Mapping files

• The word list and the mapping of each word to the sense_IDs of its corresponding senses.
• A mapping between sense_IDs and WordNet 3.0 sense keys.