In the deployment of Large Language Models (LLMs), “spurious correctness”—where answers are correct but reasoning contains errors—poses a critical risk by creating an illusion of reliability. While prior work on LLM confidence estimation focuses on answer-level or entire reasoning path confidence, these coarse-grained approaches fail to identify which specific parts of the reasoning contain errors. We propose a fine-grained confidence estimation framework that computes confidence scores for individual evidence triplets within reasoning chains, enabling precise localization of errors. Using carefully designed prompts, we generate answers, evidence in triplet format, and their respective confidence scores simultaneously, allowing automatic detection of spurious correctness patterns where partial evidence contains factual errors. Evaluated on both Japanese and English multi-hop QA benchmarks across multiple models from three model families representing different architectures and training approaches, we show that our approach exhibits superior calibration performance for evidence confidence and demonstrates effective ability to detect spurious correct answers (up to 0.84 on our primary discrimination metric). The consistent improvements across languages demonstrate the generalizability of our method. As a secondary benefit, joint generation of confidence scores improves answer confidence calibration by up to 43%. This prompt-based approach requires no model retraining and is immediately applicable to existing LLMs.

Recent LLMs show an impressive accuracy on one of the hallmark tasks of language understanding, namely Question Answering (QA). However, it is not clear if the correct answers provided by LLMs are actually grounded on the correct knowledge related to the question. In this paper, we use multi-hop QA datasets to evaluate the accuracy of the knowledge LLMs use to answer questions, and show that as much as 31% of the correct answers by the LLMs are in fact spurious, i.e., the knowledge LLMs used to ground the answer is wrong while the answer is correct. We present an analysis of these spurious correct answers by GPT-4 using three datasets in two languages, while suggesting future pathways to correct the grounding information using existing external knowledge bases.

We present JEMHopQA, a multi-hop QA dataset for the development of explainable QA systems. The dataset consists not only of question-answer pairs, but also of supporting evidence in the form of derivation triples, which contributes to making the QA task more realistic and difficult. It is created based on Japanese Wikipedia using both crowd-sourced human annotation as well as prompting a large language model (LLM), and contains a diverse set of question, answer and topic categories as compared with similar datasets released previously. We describe the details of how we built the dataset as well as the evaluation of the QA task presented by this dataset using GPT-4, and show that the dataset is sufficiently challenging for the state-of-the-art LLM while showing promise for combining such a model with existing knowledge resources to achieve better performance.

This paper describes successful applications of discriminative lexicon models to the statistical machine translation (SMT) systems into morphologically complex languages. We extend the previous work on discriminatively trained lexicon models to include more contextual information in making lexical selection decisions by building a single global log-linear model of translation selection. In offline experiments, we show that the use of the expanded contextual information, including morphological and syntactic features, help better predict words in three target languages with complex morphology (Bulgarian, Czech and Korean). We also show that these improved lexical prediction models make a positive impact in the end-to-end SMT scenario from English to these languages.

We present RefRef, a tool for viewing and exploring coreference space, which is publicly available for research purposes. Unlike similar tools currently available whose main goal is to assist the annotation process of coreference links, RefRef is dedicated for viewing and exploring coreference-annotated data, whether manually tagged or automatically resolved. RefRef is also highly customizable, as the tool is being made available with the source code. In this paper we describe the main functionalities of RefRef as well as some possibilities for customization to meet the specific needs of the users of such coreference-annotated text.

We present an automated, system-internal evaluation technique for linguistic representations in a large-scale, multilingual MT system. We use machine-learned classifiers to recognize the differences between linguistic representations generated from transfer in an MT context from representations that are produced by "native" analysis of the target language. In the MT scenario, convergence of the two is the desired result. Holding the feature set and the learning algorithm constant, the accuracy of the classifiers provides a measure of the overall difference between the two sets of linguistic representations: classifiers with higher accuracy correspond to more pronounced differences between representations. More importantly, the classifiers yield the basis for error-analysis by providing a ranking of the importance of linguistic features. The more salient a linguistic criterion is in discriminating transferred representations from "native" representations, the more work will be needed in order to get closer to the goal of producing native-like MT. We present results from using this approach on the Microsoft MT system and discuss its advantages and possible extensions.