Overview

Data

Geoquery data has been used. There are total 880 sentences-logical form pair (600 train, 280 test).

Models

Sequence-to-Sequence

Sequence-to-Sequence + Attention

Attention Weight Matrix

Sentence 1 : How high is m0?

Sentence 2 : Which state boder s0?

Base Implementation :

https://github.com/Alex-Fabbri/lang2logic-PyTorch

Original Paper:

https://arxiv.org/pdf/1601.01280.pdf