INTERACTION: A Generative XAI Framework for Natural Language Inference Explanations

Jialin Yu, Alexandra I. Cristea, Anoushka Harit, Zhongtian Sun, Olanrewaju Tahir Aduragba, Lei Shi, Noura Al Moubayed

Abstract

XAI with natural language processing aims to produce human-readable explanations as evidence for AI decision-making, which addresses explainability and transparency. However, from an HCI perspective, the current approaches only focus on delivering a single explanation, which fails to account for the diversity of human thoughts and experiences in language. This paper thus addresses this gap, by proposing a generative XAI framework, INTERACTION (explaIn aNd predicT thEn queRy with contextuAl CondiTional varIational autO-eNcoder). Our novel framework presents explanation in two steps: (step one) Explanation and Label Prediction; and (step two) Diverse Evidence Generation. We conduct intensive experiments with the Transformer architecture on a benchmark dataset, e-SNLI. Our method achieves competitive or better performance against state-of-the-art baseline models on explanation generation (up to 4.7% gain in BLEU) and prediction (up to 4.4% gain in accuracy) in step one; it can also generate multiple diverse explanations in step two.

Introduction

Deep learning models are notoriously "black boxes." While Explainable AI (XAI) aims to fix this by providing human-readable evidence for a model's decision, most current methods only provide a single explanation. From a Human-Computer Interaction (HCI) perspective, this is a flaw. Human thought is diverse; two people might explain the same conclusion in different ways based on their background, social expectations, or cognitive biases.

Furthermore, Natural Language Inference (NLI) datasets often contain "annotation artifacts"—spurious correlations that allow models to "cheat" and guess labels without truly understanding the logic. The authors argue that a robust XAI system should be able to provide a variety of valid explanations and that this process should be deeply integrated with the decision-making itself.

The INTERACTION Framework

The proposed INTERACTION framework (explaIn aNd predicT thEn queRy with contextuAl CondiTional varIational autO-eNcoder) operates in two distinct steps to bridge the gap between model transparency and human cognitive diversity.

The INTERACTION framework workflow, illustrating the two-step process from initial prediction to diverse evidence generation and potential human intervention.

Technical Methodology

  1. Explanation and Label Prediction: The model predicts the most probable label (e.g., Entailment, Contradiction, or Neutral) and a primary explanation simultaneously.
  2. Diverse Evidence Generation: Using a deep generative model, the system explores the "latent space" (a mathematical representation of possible explanations) to generate multiple, diverse alternative justifications.

A key innovation is the Contextual Convolutional Neural Encoder (Concoder). Standard Transformers produce variable-length sequences, which are difficult to map to the fixed-size latent distributions required by Variational Autoencoders. The Concoder uses 2D-convolutions (unigram, bigram, and trigram filters) to "flatten" Transformer outputs into a stable format for the generative process.

Experimental Results

The study utilizes e-SNLI (extended Stanford Natural Language Inference), a gold-standard benchmark where premise-hypothesis pairs are supplemented with human-written natural language explanations.

Performance Gains

One of the most significant findings was that models trained to generate explanations actually became better at the primary task. The INTERACTION M1 variant reached 83.42% accuracy, significantly higher than the baseline models. This suggests that the requirement to "explain itself" forces the model to learn superior semantic representations.

Baseline accuracy comparison for different encoder architectures on the NLI task, providing context for the improvements achieved by the INTERACTION framework.

Beyond accuracy, the framework achieved up to a 4.7% gain in BLEU score compared to state-of-the-art baselines. Qualitatively, the model successfully generated multiple linguistically different but semantically valid explanations for the same data point. For example, for a "Contradiction" label, it might generate: "The boy cannot be playing and sitting at the same time," and "The boy is either playing or sitting, not both."

Conclusion and Future Work

While the framework is a leap forward for diverse XAI, several areas for growth remain. The model currently uses MAP (Maximum A Posteriori) decoding, which may not always capture the most creative explanations. Additionally, the paper raises an open question for the HCI community: How many explanations are "too many"? Future work is needed to determine how users interact with multiple explanations and how they choose which one to trust.

Summary updated

Abstract

XAI with natural language processing aims to produce human-readable explanations as evidence for AI decision-making, which addresses explainability and transparency. However, from an HCI perspective, the current approaches only focus on delivering a single explanation, which fails to account for the diversity of human thoughts and experiences in language. This paper thus addresses this gap, by proposing a generative XAI framework, INTERACTION (explaIn aNd predicT thEn queRy with contextuAl CondiTional varIational autO-eNcoder). Our novel framework presents explanation in two steps: (step one) Explanation and Label Prediction; and (step two) Diverse Evidence Generation. We conduct intensive experiments with the Transformer architecture on a benchmark dataset, e-SNLI. Our method achieves competitive or better performance against state-of-the-art baseline models on explanation generation (up to 4.7% gain in BLEU) and prediction (up to 4.4% gain in accuracy) in step one; it can also generate multiple diverse explanations in step two.
BibTeX
@inproceedings{yu2022interaction,
  author = {Jialin Yu and Alexandra I. Cristea and Anoushka Harit and Zhongtian Sun and Olanrewaju Tahir Aduragba and Lei Shi and Noura Al Moubayed},
  title = {{INTERACTION: A Generative XAI Framework for Natural Language Inference Explanations}},
  booktitle = {IJCNN},
  year = {2022},
  url = {https://openreview.net/forum?id=cld-TGiLmB},
}
Figure