text2log

Neural Machine Translation (seq2seq): English => Predicate Logic

Input sentence: every cat likes fish
Decoded sentence: 'all x1.(_cat(x1) -> exists x2.(_fish(x2) & _like(x1,x2)))'

Input sentence: some people are evil and some people are good
Decoded sentence: 'exists x1.(_people(x1) & _good(x1)) & exists x2.(_people(x2) & _evil(x2))'

Input sentence: she is willing and able
Decoded sentence: 'exists x1.(_able(x1) & _willing(x1))'

Models

1. Plain LSTM
2. Plain GRU
3. Bi-directional GRU + Attention (Bahdanau, 2014)
4. Bi-directional LSTM + Attention