AMR4NLI: Interpretable and robust NLI measures from semantic graphs
This work addresses the need for more explicit and interpretable measures in NLI, which is incremental as it builds on existing semantic representations without introducing a new paradigm.
The paper tackled the problem of making natural language inference (NLI) more interpretable and robust by comparing semantic structures like contextualized embeddings and Abstract Meaning Representations (AMR) to measure entailment as a semantic substructure. The result showed that both approaches provide complementary signals and can be combined in a hybrid model, as evaluated on three English benchmarks.
The task of natural language inference (NLI) asks whether a given premise (expressed in NL) entails a given NL hypothesis. NLI benchmarks contain human ratings of entailment, but the meaning relationships driving these ratings are not formalized. Can the underlying sentence pair relationships be made more explicit in an interpretable yet robust fashion? We compare semantic structures to represent premise and hypothesis, including sets of contextualized embeddings and semantic graphs (Abstract Meaning Representations), and measure whether the hypothesis is a semantic substructure of the premise, utilizing interpretable metrics. Our evaluation on three English benchmarks finds value in both contextualized embeddings and semantic graphs; moreover, they provide complementary signals, and can be leveraged together in a hybrid model.