Seungyoon Lee, Minhyuk Kim, Seongtae Hong, Youngjoon Jang, Dongsuk Oh, Heuiseok Lim. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recently, large language models (LLMs) have made significant progress through retrieval-augmented generation (RAG) and preference learning. However, they still exhibit issues such as confirmation bias, the tendency to favor information that confirms one’s beliefs, which remains largely unexplored in current research. In this paper, we propose a novel approach to mitigate confirmation bias-induced hallucination in LLMs through a synthetic data construction pipeline and Direct Preference Optimization (DPO) training. Our method enhances the integration of diverse and complementary information from multiple passages retrieved by RAG, enabling more balanced and accurate reasoning. Experimental results demonstrate significant improvements in response accuracy and reduced hallucination on benchmarks such as Natural Questions Open and HaluBench. These findings suggest that our approach effectively mitigates confirmation bias in long-context question answering, with potential applications to other NLP tasks. We release our data, and evaluation/train code for public access.3]https://github.com/OccasionallyNLP/Synthetic-Paths-to-Integral-Truth.git

Humans usually have conversations by making use of prior knowledge about a topic and background information of the people whom they are talking to. However, existing conversational agents and datasets do not consider such comprehensive information, and thus they have a limitation in generating the utterances where the knowledge and persona are fused properly. To address this issue, we introduce a call For Customized conversation (FoCus) dataset where the customized answers are built with the user's persona and Wikipedia knowledge. To evaluate the abilities to make informative and customized utterances of pre-trained language models, we utilize BART and GPT-2 as well as transformer-based models. We assess their generation abilities with automatic scores and conduct human evaluations for qualitative results. We examine whether the model reflects adequate persona and knowledge with our proposed two sub-tasks, persona grounding (PG) and knowledge grounding (KG). Moreover, we show that the utterances of our data are constructed with the proper knowledge and persona through grounding quality assessment.

Recent pre-trained language models (PLMs) achieved great success on many natural language processing tasks through learning linguistic features and contextualized sentence representation. Since attributes captured in stacked layers of PLMs are not clearly identified, straightforward approaches such as embedding the last layer are commonly preferred to derive sentence representations from PLMs. This paper introduces the attention-based pooling strategy, which enables the model to preserve layer-wise signals captured in each layer and learn digested linguistic features for downstream tasks. The contrastive learning objective can adapt the layer-wise attention pooling to both unsupervised and supervised manners. It results in regularizing the anisotropic space of pre-trained embeddings and being more uniform. We evaluate our model on standard semantic textual similarity (STS) and semantic search tasks. As a result, our method improved the performance of the base contrastive learned BERT_{base} and variants.

In this paper, we study the task of selecting the optimal response given a user and system utterance history in retrieval-based multi-turn dialog systems. Recently, pre-trained language models (e.g., BERT, RoBERTa, and ELECTRA) showed significant improvements in various natural language processing tasks. This and similar response selection tasks can also be solved using such language models by formulating the tasks as dialog--response binary classification tasks. Although existing works using this approach successfully obtained state-of-the-art results, we observe that language models trained in this manner tend to make predictions based on the relatedness of history and candidates, ignoring the sequential nature of multi-turn dialog systems. This suggests that the response selection task alone is insufficient for learning temporal dependencies between utterances. To this end, we propose utterance manipulation strategies (UMS) to address this problem. Specifically, UMS consist of several strategies (i.e., insertion, deletion, and search), which aid the response selection model towards maintaining dialog coherence. Further, UMS are self-supervised methods that do not require additional annotation and thus can be easily incorporated into existing approaches. Extensive evaluation across multiple languages and models shows that UMS are highly effective in teaching dialog consistency, which leads to models pushing the state-of-the-art with significant margins on multiple public benchmark datasets.

CommonsenseQA is a task in which a correct answer is predicted through commonsense reasoning with pre-defined knowledge.Most previous works have aimed to improve the performance with distributed representation without considering the process of predicting the answer from the semantic representation of the question.To shed light upon the semantic interpretation of the question, we propose an AMR-ConceptNet-Pruned (ACP) graph.The ACP graph is pruned from a full integrated graph encompassing Abstract Meaning Representation (AMR) graph generated from input questions and an external commonsense knowledge graph, ConceptNet (CN).Then the ACP graph is exploited to interpret the reasoning path as well as to predict the correct answer on the CommonsenseQA task.This paper presents the manner in which the commonsense reasoning process can be interpreted with the relations and concepts provided by the ACP graph.Moreover, ACP-based models are shown to outperform the baselines.

We focus on multi-turn response selection in a retrieval-based dialog system. In this paper, we utilize the powerful pre-trained language model Bi-directional Encoder Representations from Transformer (BERT) for a multi-turn dialog system and propose a highly effective post-training method on domain-specific corpus. Although BERT is easily adopted to various NLP tasks and outperforms previous baselines of each task, it still has limitations if a task corpus is too focused on a certain domain. Post-training on domain-specific corpus (e.g., Ubuntu Corpus) helps the model to train contextualized representations and words that do not appear in general corpus (e.g., English Wikipedia). Experimental results show that our approach achieves new state-of-the-art on two response selection benchmarks (i.e., Ubuntu Corpus V1, Advising Corpus) performance improvement by 5.9% and 6% on R@1.

Word sense disambiguation (WSD) is the task to determine the word sense according to its context. Many existing WSD studies have been using an external knowledge-based unsupervised approach because it has fewer word set constraints than supervised approaches requiring training data. In this paper, we propose a new WSD method to generate the context of an ambiguous word by using similarities between an ambiguous word and words in the input document. In addition, to leverage our WSD method, we further propose a new word similarity calculation method based on the semantic network structure of BabelNet. We evaluate the proposed methods on the SemEval-13 and SemEval-15 for English WSD dataset. Experimental results demonstrate that the proposed WSD method significantly improves the baseline WSD method. Furthermore, our WSD system outperforms the state-of-the-art WSD systems in the Semeval-13 dataset. Finally, it has higher performance than the state-of-the-art unsupervised knowledge-based WSD system in the average performance of both datasets.