Large Language Models (LLMs) excel at understanding the semantic relationships between queries and documents, even with lengthy and complex long-tail queries. These queries are challenging for feedback-based rankings due to sparse user engagement and limited feedback, making LLMs’ ranking ability highly valuable. However, the large size and slow inference of LLMs necessitate the development of smaller, more efficient models (sLLMs). Recently, integrating ranking label generation into distillation techniques has become crucial, but existing methods underutilize LLMs’ capabilities and are cumbersome. Our research, RRADistill: Re-Ranking Ability Distillation, propose an efficient label generation pipeline and novel sLLM training methods for both encoder and decoder models. We introduce an encoder-based method using a Term Control Layer to capture term matching signals and a decoder-based model with a ranking layer for enhanced understanding. A/B testing on a Korean-based search platform, validates the effectiveness of our approach in improving re-ranking for long-tail queries.

Most prior safety research of large language models (LLMs) has focused on enhancing the alignment of LLMs to better suit the safety requirements of their use cases. However, internalizing such safeguard features into larger models brought challenges of higher training cost and unintended degradation of helpfulness. In this paper, we leverage a smaller LLM for both harmful query detection and safeguard response generation. We introduce our safety requirements and the taxonomy of harmfulness categories, and then propose a multi-task learning mechanism fusing the two tasks into a single model. We demonstrate the effectiveness of our approach, providing on par or surpassing harmful query detection and safeguard response performance compared to the publicly available LLMs.

This paper presents a English-Korean parallel dataset that collects 381K news articles where 1,400 of them, comprising 10K sentences, are manually labeled for crosslingual named entity recognition (NER). The annotation guidelines for the two languages are developed in parallel, that yield the inter-annotator agreement scores of 91 and 88% for English and Korean respectively, indicating sublime quality annotation in our dataset. Three types of crosslingual learning approaches, direct model transfer, embedding projection, and annotation projection, are used to develop zero-shot Korean NER models. Our best model gives the F1-score of 51% that is very encouraging, considering the extremely distinct natures of these two languages. This is pioneering work that explores zero-shot cross-lingual learning between English and Korean and provides rich parallel annotation for a core NLP task such as named entity recognition.

This paper presents a new corpus and annotation guideline for a novel coreference resolution task on fictional texts, and analyzes its unique characteristics. FantasyCoref contains 211 stories of Grimms’ Fairy Tales and 3 other fantasy literature annotated in the omniscient writer’s point of view (OWV) to handle distinctive aspects in this genre. This task is more challenging than general coreference resolution in two ways. First, documents in our corpus are 2.5 times longer than the ones in OntoNotes, raising a new layer of difficulty in resolving long-distant referents. Second, annotation of literary styles and concepts raise several issues which are not sufficiently addressed in the existing annotation guidelines. Hence, considerations on such issues and the concept of OWV are necessary to achieve high inter-annotator agreement (IAA) in coreference resolution of fictional texts. We carefully conduct annotation tasks in four stages to ensure the quality of our annotation. As a result, a high IAA score of 87% is achieved using the standard coreference evaluation metric. Finally, state-of-the-art coreference resolution approaches are evaluated on our corpus. After training with our annotated dataset, there was a 2.59% and 3.06% improvement over the model trained on the OntoNotes dataset. Also, we observe that the portion of errors specific to fictional texts declines after the training.