SLITE: An Interpretable Hybrid Model for Textual Entailment Using Linguistic Features

Forum topic · 小凯 · 2026-09-23

Summary

Researchers David Torres-Moreno, Jorge Hermosillo-Valadez, and Asela Reig-Alamillo present SLITE, an explainable hybrid model for Recognizing Textual Entailment (RTE) on arXiv (2609.24932). The model combines two complementary semantic layers: a structural-relational layer based on semantic compatibility between compositional entities, and a distributional-informational layer based on patterns of information change between premise and hypothesis embeddings. Seventeen features combine entity-level semantic relations, polarity-sensitive lexical matching, and alignment measures (including entropy and transfer entropy) over sub-representations of similarity matrices. A logistic regression model trained on these features achieves 83% accuracy on three-class SICK and 96% on SICK-CE, outperforming IsoLex by four points and trailing RoBERTa by less than two points at a fraction of the computational cost. Ablation studies and SHAP analysis show structural-relational features drive classification, while distributional-informational features add complementary value, especially for detecting neutral and contradictory relations. The authors argue hybrid approaches are a viable, interpretable alternative to large neural architectures.

Overview

Research area: NLP Authors: David Torres-Moreno, Jorge Hermosillo-Valadez, Asela Reig-Alamillo Published: 2026-09-21 arXiv: 2609.24932

Abstract

Despite the success of neural models in natural language processing, their black-box nature limits interpretability and conceals the linguistic phenomena underlying their predictions. The authors present SLITE, an explainable hybrid model for Recognizing Textual Entailment that integrates two complementary layers of semantic analysis:

A structural-relational layer, based on semantic compatibility and incompatibility between compositional entities.
A distributional-informational layer, based on structured patterns of information change between embedding-based representations of the premise and the hypothesis.

They propose 17 features combining entity-level semantic relations, polarity-sensitive lexical matching, and alignment measures over semantic sub-representations of the similarity matrix, including entropy- and transfer-entropy-based measures.

Key Results

A logistic regression model trained on these features reaches 83% accuracy on three-class SICK and 96% on SICK-CE.
It outperforms IsoLex by 4 percentage points and stays within 2 points of RoBERTa, at a small fraction of the computational cost.
Ablation studies and SHAP analysis confirm that structural-relational features are the primary drivers of classification, while distributional-informational features provide complementary contributions, particularly for detecting neutral and contradiction relations.

Conclusion

The results suggest that hybrid interpretable methods are a viable and scientifically productive alternative to large-scale neural architectures, strengthening the dialogue between linguistic theory and computational modeling of inference.

--- *Collected automatically on 2026-09-23.*