QLSC: A Query Latent Semantic Calibrator for Robust Extractive Question Answering

TL;DR

The paper tackles the challenge of robustness in extractive QA when questions are paraphrase variants of the same meaning. It introduces the Query Latent Semantic Calibrator (QLSC), a plug-in module that learns latent semantic center features via Semantic Center Learning, Soft Semantic Feature Selection, and Query Semantic Calibration, integrating these centers into query and passage embeddings through attention. The approach uses a Subspace Mapping Network to project information and query vectors into multiple subspaces, fusion to form latent centers, and an attention-based calibration to align paraphrase variations with the relevant passages, all optimized with standard QA losses. Empirical results on Dureader$_{robust}$ and SQuAD1.1 demonstrate improved robustness across over-sensitivity, over-stability, and generalization dimensions, with consistent gains across multiple backbones and setups, and ablations validate the effectiveness of the subspace count, feature selection, and calibration mechanisms. The work offers a practical, encoder-agnostic augmentation that enhances semantic alignment between queries and passages, enabling more reliable extractive QA in real-world, paraphrase-rich scenarios.

Abstract

Extractive Question Answering (EQA) in Machine Reading Comprehension (MRC) often faces the challenge of dealing with semantically identical but format-variant inputs. Our work introduces a novel approach, called the ``Query Latent Semantic Calibrator (QLSC)'', designed as an auxiliary module for existing MRC models. We propose a unique scaling strategy to capture latent semantic center features of queries. These features are then seamlessly integrated into traditional query and passage embeddings using an attention mechanism. By deepening the comprehension of the semantic queries-passage relationship, our approach diminishes sensitivity to variations in text format and boosts the model's capability in pinpointing accurate answers. Experimental results on robust Question-Answer datasets confirm that our approach effectively handles format-variant but semantically identical queries, highlighting the effectiveness and adaptability of our proposed method.

Figures (5)

• Figure 1: The architecture of our proposed QLSC method. SCL represents Semantic Center Learning, SSFS represents Soft Semantic Feature Selection, and QSC represents Query Semantic Calibration. $C$ is the information matrix and $m$ is the number of subspaces. $k$ is the number of the semantic center feature $T$. $l_{q}$ and $l_{p}$ respectively represent the length of the query and the passage encoded by the pre-trained language model.
• Figure 2: The effect of a different number of information features on model performance on Dev set.
• Figure 3: The influence of various random seeds on model performance.
• Figure 4: The training loss of different models under each epochs.
• Figure 5: PCA dimensionality reduction visualization results for querying embedded vectors.