Consistency-based methods have emerged as an effective approach to uncertainty quantification (UQ) in large language models. These methods typically rely on several generations obtained via multinomial sampling, measuring their agreement level. However, in short-form QA, multinomial sampling is prone to producing duplicates due to peaked distributions, and its stochasticity introduces considerable variance in uncertainty estimates across runs. We introduce a new family of methods that employ beam search to generate candidates for consistency-based UQ, yielding improved performance and reduced variance compared to multinomial sampling. We also provide a theoretical lower bound on the beam set probability mass under which beam search achieves a smaller error than multinomial sampling. We empirically evaluate our approach on six QA datasets and find that its consistent improvements over multinomial sampling lead to state-of-the-art UQ performance.

Ensuring the reliability and safety of machine learning models in open-world deployment is a central challenge in AI safety. This thesis develops both algorithmic and theoretical foundations to address key reliability issues arising from distributional uncertainty and unknown classes, from standard neural networks to modern foundation models like large language models (LLMs). Traditional learning paradigms, such as empirical risk minimization (ERM), assume no distribution shift between training and inference, often leading to overconfident predictions on out-of-distribution (OOD) inputs. This thesis introduces novel frameworks that jointly optimize for in-distribution accuracy and reliability to unseen data. A core contribution is the development of an unknown-aware learning framework that enables models to recognize and handle novel inputs without labeled OOD data. We propose new outlier synthesis methods, VOS, NPOS, and DREAM-OOD, to generate informative unknowns during training. Building on this, we present SAL, a theoretical and algorithmic framework that leverages unlabeled in-the-wild data to enhance OOD detection under realistic deployment conditions. These methods demonstrate that abundant unlabeled data can be harnessed to recognize and adapt to unforeseen inputs, providing formal reliability guarantees. The thesis also extends reliable learning to foundation models. We develop HaloScope for hallucination detection in LLMs, MLLMGuard for defending against malicious prompts in multimodal models, and data cleaning methods to denoise human feedback used for better alignment. These tools target failure modes that threaten the safety of large-scale models in deployment. Overall, these contributions promote unknown-aware learning as a new paradigm, and we hope it can advance the reliability of AI systems with minimal human efforts.

As the legal community increasingly examines the use of large language models (LLMs) for various legal applications, legal AI developers have turned to retrieval-augmented LLMs ("RAG" systems) to improve system performance and robustness. An obstacle to the development of specialized RAG systems is the lack of realistic legal RAG benchmarks which capture the complexity of both legal retrieval and downstream legal question-answering. To address this, we introduce two novel legal RAG benchmarks: Bar Exam QA and Housing Statute QA. Our tasks correspond to real-world legal research tasks, and were produced through annotation processes which resemble legal research. We describe the construction of these benchmarks and the performance of existing retriever pipelines. Our results suggest that legal RAG remains a challenging application, thus motivating future research.

This paper presents a method and supporting tools for automation of terminology management for Arabic. The tools extract lists of parallel terminology matching terms in foreign languages to their Arabic counterparts from field specific texts. This has significant implications as it can be used to improve consistent translation and use of terms in specialized Arabic academic books, and provides automated aid for enhancing cross lingual text processing. This automation of terminology management aims to reduce processing time, and ensure use of consistent and correct terminology. The extraction takes advantage of naturally occurring term translations. It considers several candidate phrases of varying lengths that co-occur next to the foreign terms. Then it computes several similarity metrics, including lexicographic, phonetic, morphological, and semantic ones to decide the problem. We experiment with heuristic, machine learning, and ML with post processing approaches. This paper reports on a novel curated dataset for the task, an existing expert reviewed industry parallel corpora, and on the performance of the three approaches. The best approach achieved 94.9% precision and 92.4% recall.

Back-of-the-book indexes are crucial for book readability. Their manual creation is laborious and error prone. In this paper, we consider automating back-of-the-book index extraction for Arabic books to help simplify both the creation and review tasks. Given a back-of-the-book index, we aim to check and identify the accurate occurrences of index terms relative to the associated pages. To achieve this, we first define a pool of candidates for each term by extracting all possible noun phrases from paragraphs appearing on the relevant index pages. These noun phrases, identified through part-of-speech analysis, are stored in a vector database for efficient retrieval. We use several metrics, including exact matches, lexical similarity, and semantic similarity, to determine the most appropriate occurrence. The candidate with the highest score based on these metrics is chosen as the occurrence of the term. We fine-tuned a heuristic method, that considers the above metrics and that achieves an F1-score of .966 (precision=.966, recall=.966). These excellent results open the door for future work related to automation of back-of-the-book index generation and checking.

Clinical question answering systems have the potential to provide clinicians with relevant and timely answers to their questions. Nonetheless, despite the advances that have been made, adoption of these systems in clinical settings has been slow. One issue is a lack of question-answering datasets which reflect the real-world needs of health professionals. In this work, we present RealMedQA, a dataset of realistic clinical questions generated by humans and an LLM. We describe the process for generating and verifying the QA pairs and assess several QA models on BioASQ and RealMedQA to assess the relative difficulty of matching answers to questions. We show that the LLM is more cost-efficient for generating "ideal" QA pairs. Additionally, we achieve a lower lexical similarity between questions and answers than BioASQ which provides an additional challenge to the top two QA models, as per the results. Introduction Clinical question answering (QA) systems could allow clinicians to find timely and relevant answers to questions occurring during consultations in real-time [1, 2, 3, 4, 5].

Textual noise, such as typos or abbreviations, is a well-known issue that penalizes vanilla Transformers for most downstream tasks. We show that this is also the case for sentence similarity, a fundamental task in multiple domains, e.g. matching, retrieval or paraphrasing. Sentence similarity can be approached using cross-encoders, where the two sentences are concatenated in the input allowing the model to exploit the inter-relations between them. Previous works addressing the noise issue mainly rely on data augmentation strategies, showing improved robustness when dealing with corrupted samples that are similar to the ones used for training. However, all these methods still suffer from the token distribution shift induced by typos. In this work, we propose to tackle textual noise by equipping cross-encoders with a novel LExical-aware Attention module (LEA) that incorporates lexical similarities between words in both sentences. By using raw text similarities, our approach avoids the tokenization shift problem obtaining improved robustness. We demonstrate that the attention bias introduced by LEA helps cross-encoders to tackle complex scenarios with textual noise, specially in domains with short-text descriptions and limited context. Experiments using three popular Transformer encoders in five e-commerce datasets for product matching show that LEA consistently boosts performance under the presence of noise, while remaining competitive on the original (clean) splits. We also evaluate our approach in two datasets for textual entailment and paraphrasing showing that LEA is robust to typos in domains with longer sentences and more natural context. Additionally, we thoroughly analyze several design choices in our approach, providing insights about the impact of the decisions made and fostering future research in cross-encoders dealing with typos.

While multilingual neural machine translation has achieved great success, it suffers from the off-target issue, where the translation is in the wrong language. This problem is more pronounced on zero-shot translation tasks. In this work, we find that failing in encoding discriminative target language signal will lead to off-target and a closer lexical distance (i.e., KL-divergence) between two languages' vocabularies is related with a higher off-target rate. We also find that solely isolating the vocab of different languages in the decoder can alleviate the problem. Motivated by the findings, we propose Language Aware Vocabulary Sharing (LAVS), a simple and effective algorithm to construct the multilingual vocabulary, that greatly alleviates the off-target problem of the translation model by increasing the KL-divergence between languages. We conduct experiments on a multilingual machine translation benchmark in 11 languages. Experiments show that the off-target rate for 90 translation tasks is reduced from 29\% to 8\%, while the overall BLEU score is improved by an average of 1.9 points without extra training cost or sacrificing the supervised directions' performance. We release the code at https://github.com/PKUnlp-icler/Off-Target-MNMT for reproduction.

Retrieval-enhanced language models (LMs), which condition their predictions on text retrieved from large external datastores, have recently shown significant perplexity improvements compared to standard LMs. One such approach, the $k$NN-LM, interpolates any existing LM's predictions with the output of a $k$-nearest neighbors model and requires no additional training. In this paper, we explore the importance of lexical and semantic matching in the context of items retrieved by $k$NN-LM. We find two trends: (1) the presence of large overlapping $n$-grams between the datastore and evaluation set plays an important factor in strong performance, even when the datastore is derived from the training data; and (2) the $k$NN-LM is most beneficial when retrieved items have high semantic similarity with the query. Based on our analysis, we define a new formulation of the $k$NN-LM that uses retrieval quality to assign the interpolation coefficient. We empirically measure the effectiveness of our approach on two English language modeling datasets, Wikitext-103 and PG-19. Our re-formulation of the $k$NN-LM is beneficial in both cases, and leads to nearly 4% improvement in perplexity on the Wikitext-103 test set.

In this article I will go over the intuition behind how Levenshtein distance works and how to use Levenshtein distance in building a plagiarism detection pipeline. Identifying similarity between text is a common problem in NLP and is used by many companies world wide. The most common application of text similarity comes from the form of identifying plagiarized text. Educational facilities ranging from elementary school, high school, college and universities all around the world use services like Turnitin to ensure that the work submitted by students is original and their own. Other applications of text similarity is commonly used by companies which have a similar structure to Stack Overflow or Stack Exchange.