Estimating the confidence of large language model (LLM) outputs is essential for real-world applications requiring high user trust. Black-box uncertainty quantification (UQ) methods, relying solely on model API access, have gained popularity due to their practical benefits. In this paper, we examine the implicit assumption behind several UQ methods, which use generation consistency as a proxy for confidence-an idea we formalize as the consistency hypothesis. We introduce three mathematical statements with corresponding statistical tests to capture variations of this hypothesis and metrics to evaluate LLM output conformity across tasks. Our empirical investigation, spanning 8 benchmark datasets and 3 tasks (question answering, text summarization, and text-to-SQL), highlights the prevalence of the hypothesis under different settings. Among the statements, we highlight the ‘Sim-Any’ hypothesis as the most actionable, and demonstrate how it can be leveraged by proposing data-free black-box UQ methods that aggregate similarities between generations for confidence estimation. These approaches can outperform the closest baselines, showcasing the practical value of the empirically observed consistency hypothesis.

Michael Glass, Gaetano Rossiello, Md Faisal Mahbub Chowdhury, Ankita Naik, Pengshan Cai, Alfio Gliozzo. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Automatically inducing high quality knowledge graphs from a given collection of documents still remains a challenging problem in AI.One way to make headway for this problem is through advancements in a related task known as slot filling.In this task, given an entity query in form of [ENTITY, SLOT, ?], a system is asked to 'fill' the slot by generating or extracting the missing value exploiting evidence extracted from relevant passage(s) in the given document collection.The recent works in the field try to solve this task in an end-to-end fashion using retrieval-based language models.In this paper, we present a novel approach to zero-shot slot filling that extends dense passage retrieval with hard negatives and robust training procedures for retrieval augmented generation models.Our model reports large improvements on both T-REx and zsRE slot filling datasets, improving both passage retrieval and slot value generation, and ranking at the top-1 position in the KILT leaderboard.Moreover, we demonstrate the robustness of our system showing its domain adaptation capability on a new variant of the TACRED dataset for slot filling, through a combination of zero/few-shot learning.We release the source code and pre-trained models 1 .

Michael Glass, Mustafa Canim, Alfio Gliozzo, Saneem Chemmengath, Vishwajeet Kumar, Rishav Chakravarti, Avi Sil, Feifei Pan, Samarth Bharadwaj, Nicolas Rodolfo Fauceglia. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Research on the task of Reading Comprehension style Question Answering (RCQA) has gained momentum in recent years due to the emergence of human annotated datasets and associated leaderboards, for example CoQA, HotpotQA, SQuAD, TriviaQA, etc. While state-of-the-art has advanced considerably, there is still ample opportunity to advance it further on some important variants of the RCQA task. In this paper, we propose a novel deep neural architecture, called TAP (Translucent Answer Prediction), to identify answers and evidence (in the form of supporting facts) in an RCQA task requiring multi-hop reasoning. TAP comprises two loosely coupled networks – Local and Global Interaction eXtractor (LoGIX) and Answer Predictor (AP). LoGIX predicts supporting facts, whereas AP consumes these predicted supporting facts to predict the answer span. The novel design of LoGIX is inspired by two key design desiderata – local context and global interaction– that we identified by analyzing examples of multi-hop RCQA task. The loose coupling between LoGIX and the AP reveals the set of sentences used by the AP in predicting an answer. Therefore, answer predictions of TAP can be interpreted in a translucent manner. TAP offers state-of-the-art performance on the HotpotQA (Yang et al. 2018) dataset – an apt dataset for multi-hop RCQA task – as it occupies Rank-1 on its leaderboard (https://hotpotqa.github.io/) at the time of submission.

Michael Glass, Alfio Gliozzo, Rishav Chakravarti, Anthony Ferritto, Lin Pan, G P Shrivatsa Bhargav, Dinesh Garg, Avi Sil. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020.

We propose a novel approach to learn representations of relations expressed by their textual mentions. In our assumption, if two pairs of entities belong to the same relation, then those two pairs are analogous. We collect a large set of analogous pairs by matching triples in knowledge bases with web-scale corpora through distant supervision. This dataset is adopted to train a hierarchical siamese network in order to learn entity-entity embeddings which encode relational information through the different linguistic paraphrasing expressing the same relation. The model can be used to generate pre-trained embeddings which provide a valuable signal when integrated into an existing neural-based model by outperforming the state-of-the-art methods on a relation extraction task.

We introduce Grinch, a new algorithm for large-scale, non-greedy hierarchical clustering with general linkage functions that compute arbitrary similarity between two point sets. The key components of Grinch are its rotate and graft subroutines that efficiently reconfigure the hierarchy as new points arrive, supporting discovery of clusters with complex structure. Grinch is motivated by a new notion of separability for clustering with linkage functions: we prove that when the linkage function is consistent with a ground-truth clustering, Grinch is guaranteed to produce a cluster tree containing the ground-truth, independent of data arrival order. Our empirical results on benchmark and author coreference datasets (with standard and learned linkage functions) show that Grinch is more accurate than other scalable methods, and orders of magnitude faster than hierarchical agglomerative clustering.

State-of-the-art relation extraction ap-proaches are only able to recognize relationships between mentions of entity arguments stated explicitly in the text and typically localized to the same sentence.However, the vast majority of relations are either implicit or not sententially localized.This is a major problem for Knowledge Base Population, severely limiting recall.In this paper we propose a new methodology to identify relations between two entities, consisting of detecting a very large number of unary relations, and using them to infer missing entities.We describe a deep learning architecture able to learn thousands of such relations very efficiently by using a common deep learning based representation.Our approach largely outperforms state of the art relation extraction technology on a newly introduced web scale knowledge base population benchmark, that we release to the research community.

We apply a general deep learning framework to address the non-factoid question answering task. Our approach does not rely on any linguistic tools and can be applied to different languages or domains. Various architectures are presented and compared. We create and release a QA corpus and setup a new QA task in the insurance domain. Experimental results demonstrate superior performance compared to the baseline methods and various technologies give further improvements. For this highly challenging task, the top-1 accuracy can reach up to 65.3% on a test set, which indicates a great potential for practical use.

In this paper we examine the lexical substitu-tion task for the medical domain. We adapt the current best system from the open domain, which trains a single classifier for all instances using delexicalized features. We show sig-nificant improvements over a strong baseline coming from a distributional thesaurus (DT). Whereas in the open domain system, features derived from WordNet show only slight im-provements, we show that its counterpart for the medical domain (UMLS) shows a signif-icant additional benefit when used for feature generation. 1

Many forms of word relatedness have been developed, providing different perspec-tives on word similarity. We introduce a Bayesian probabilistic tensor factoriza-tion model for synthesizing a single word vector representation and per-perspective linear transformations from any number of word similarity matrices. The result-ing word vectors, when combined with the per-perspective linear transformation, ap-proximately recreate while also regulariz-ing and generalizing, each word similarity perspective. Our method can combine manually cre-ated semantic resources with neural word embeddings to separate synonyms and antonyms, and is capable of generaliz-ing to words outside the vocabulary of any particular perspective. We evaluated the word embeddings with GRE antonym questions, the result achieves the state-of-the-art performance. 1