MonaLog: a Lightweight System for Natural Language Inference Based on Monotonicity
- Hai Hu (Indiana University Bloomington)
- Qi Chen (Indiana University Bloomington)
- Kyle Richardson (Allen Institute for Artificial Intelligence)
- Atreyee Mukherjee (Indiana University Bloomington)
- Lawrence S Moss (Indiana University Bloomington)
- Sandra Kübler (Indiana University Bloomington)
Abstract
We present a new logic-based inference engine for natural language inference called MonaLog, which is based on the monotonicity calculus and natural logic. In contrast to existing logic-based approaches, our system is lightweight, and operates using a small set of well-known monotonicity facts about quantifiers and lexical items. Despite its simplicity, we find it competitive with other logic-based NLI models on the SICK benchmark. We also use MonaLog incombination with BERT in a variety of settings, including data augmentation. We show that MonaLog is capable of generating large amounts of high-quality training data for BERT, improving its accuracy on SICK.
Keywords: natural language inference, monotonicity, natural logic, BERT, data augmentation
How to Cite:
Hu, H., Chen, Q., Richardson, K., Mukherjee, A., Moss, L. S. & Kübler, S., (2020) “MonaLog: a Lightweight System for Natural Language Inference Based on Monotonicity”, Society for Computation in Linguistics 3(1), 319-329. doi: https://doi.org/10.7275/te06-g113
Downloads:
Download PDF