• REFS:
• Type Labels

EXPM: Take dot product of inputs after first embedding

DESC: due to the difficulty of representing multi-token words as vectors, we only use single token words

RESULTS:

• semantically similar inputs seem to have high similarity
• tokens with spaces in front seem to have lower similarity with each other than tokens without spaces in front

TYPES: [T1]

REFS: [1]

https://colab.research.google.com/drive/1rch6VaG9O1YFJT1wPjjbXyDgXizGT7WV#scrollTo=dDRC2P8Vpk-T&line=1&uniqifier=1

<<<

EXPM: try to compare multi-token words by dot product by taking mean

RESULT: does not work

REFS: [1]

<<<