Mapping Natural-language Problems to Formal-language Solutions Using Structured Neural Representations
This addresses the challenge of mapping natural-language problems to formal-language solutions, which is incremental as it builds on existing encoder-decoder methods with structured representations.
The paper tackles the problem of generating formal-language programs from natural-language descriptions by proposing TP-N2F, a model using Tensor Product Representations to explicitly capture symbolic structure, which outperforms LSTM-based seq2seq models on two benchmarks and sets new state-of-the-art results.
Generating formal-language programs represented by relational tuples, such as Lisp programs or mathematical operations, to solve problems stated in natural language is a challenging task because it requires explicitly capturing discrete symbolic structural information implicit in the input. However, most general neural sequence models do not explicitly capture such structural information, limiting their performance on these tasks. In this paper, we propose a new encoder-decoder model based on a structured neural representation, Tensor Product Representations (TPRs), for mapping Natural-language problems to Formal-language solutions, called TP-N2F. The encoder of TP-N2F employs TPR `binding' to encode natural-language symbolic structure in vector space and the decoder uses TPR `unbinding' to generate, in symbolic space, a sequential program represented by relational tuples, each consisting of a relation (or operation) and a number of arguments. TP-N2F considerably outperforms LSTM-based seq2seq models on two benchmarks and creates new state-of-the-art results. Ablation studies show that improvements can be attributed to the use of structured TPRs explicitly in both the encoder and decoder. Analysis of the learned structures shows how TPRs enhance the interpretability of TP-N2F.