Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SLFNet: Generating Semantic Logic Forms from Natural Language Using Semantic Probability Graphs

Hao Wu, Fan Xu

TL;DR

The paper tackles the NL-to-SLF parsing problem, which is hampered by order-sensitivity in traditional seq2seq models. It introduces SLFNet, a model that fuses dependent syntactic information via a Dependency-Fused BiLSTM, and uses Semantic Probability Graphs to encode local dependencies among predictor variables, enabling Seq-to-Slots predictions with Multi-Head SLF Attention. The task is decomposed into predicting the number of SLF groups, filling slot values, and performing a final logical inference to assemble the SLFs. Empirical results show state-of-the-art performance on ChineseQCI-TS and Okapi, with competitive results on ATIS/WikiSQL, highlighting improved NL-to-SLF generation and potential for robust natural language interfaces in data-driven systems.

Abstract

Building natural language interfaces typically uses a semantic parser to parse the user's natural language and convert it into structured \textbf{S}emantic \textbf{L}ogic \textbf{F}orms (SLFs). The mainstream approach is to adopt a sequence-to-sequence framework, which requires that natural language commands and SLFs must be represented serially. Since a single natural language may have multiple SLFs or multiple natural language commands may have the same SLF, training a sequence-to-sequence model is sensitive to the choice among them, a phenomenon recorded as "order matters". To solve this problem, we propose a novel neural network, SLFNet, which firstly incorporates dependent syntactic information as prior knowledge and can capture the long-range interactions between contextual information and words. Secondly construct semantic probability graphs to obtain local dependencies between predictor variables. Finally we propose the Multi-Head SLF Attention mechanism to synthesize SLFs from natural language commands based on Sequence-to-Slots. Experiments show that SLFNet achieves state-of-the-art performance on the ChineseQCI-TS and Okapi datasets, and competitive performance on the ATIS dataset.

SLFNet: Generating Semantic Logic Forms from Natural Language Using Semantic Probability Graphs

TL;DR

The paper tackles the NL-to-SLF parsing problem, which is hampered by order-sensitivity in traditional seq2seq models. It introduces SLFNet, a model that fuses dependent syntactic information via a Dependency-Fused BiLSTM, and uses Semantic Probability Graphs to encode local dependencies among predictor variables, enabling Seq-to-Slots predictions with Multi-Head SLF Attention. The task is decomposed into predicting the number of SLF groups, filling slot values, and performing a final logical inference to assemble the SLFs. Empirical results show state-of-the-art performance on ChineseQCI-TS and Okapi, with competitive results on ATIS/WikiSQL, highlighting improved NL-to-SLF generation and potential for robust natural language interfaces in data-driven systems.

Abstract

Building natural language interfaces typically uses a semantic parser to parse the user's natural language and convert it into structured \textbf{S}emantic \textbf{L}ogic \textbf{F}orms (SLFs). The mainstream approach is to adopt a sequence-to-sequence framework, which requires that natural language commands and SLFs must be represented serially. Since a single natural language may have multiple SLFs or multiple natural language commands may have the same SLF, training a sequence-to-sequence model is sensitive to the choice among them, a phenomenon recorded as "order matters". To solve this problem, we propose a novel neural network, SLFNet, which firstly incorporates dependent syntactic information as prior knowledge and can capture the long-range interactions between contextual information and words. Secondly construct semantic probability graphs to obtain local dependencies between predictor variables. Finally we propose the Multi-Head SLF Attention mechanism to synthesize SLFs from natural language commands based on Sequence-to-Slots. Experiments show that SLFNet achieves state-of-the-art performance on the ChineseQCI-TS and Okapi datasets, and competitive performance on the ATIS dataset.
Paper Structure (14 sections, 11 equations, 3 figures, 1 table)

This paper contains 14 sections, 11 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Semantic Probability Graphs
  3. SLFNet
  4. Seq-to-Slots based prediction method using Multi-Head SLF Attention
  5. Slot values prediction
  6. Experiment
  7. Data
  8. Baselines
  9. Results and Discussion
  10. Conclusion
  11. Example Appendix
  12. The derivation process of Denpendency-Fused BiLSTM
  13. Metrics
  14. Training details

Figures (3)

  • Figure 1: A natural language example from the dataset ChineseqCI-TS. In the framework of semantic parsing of natural language commands, semantic slots are defined as three types "$ALO$", where "$A$" represents the $action\_name$ slot; "$L$" stands for slot $location\_name$; "$O$" indicates the $object\_name$ slot; The $k$ at the end of the slot name represents the number of "$ALO$" groups that the natural language command may contain. For values predictions, this is a sequence labeling problem, where lower-case "$o$" means that the token does not belong to either of the "$ALO$" slots, where "turn on" fills in the $action\_name\_1$ slot, and the prediction details are described in the model description.
  • Figure 2: SLFNet neural network architecture diagram includes three main modules: (a). Encoder-Weight calculation module; (b). Denpendency-Fused BiLSTM module; (c). Prediction Module.
  • Figure 3: Accuracy of different slot values under SLFNet neural network.