Publications

2025

1. Trustworthy AI

   CiteEval: Principle-Driven Citation Evaluation for Source Attribution

   Yumo Xu, Peng Qi*, Jifan Chen*, Kunlun Liu, Rujun Han, Lan Liu, Bonan Min, Vittorio Castelli, Arshit Gupta, and Zhiguo Wang

   In ACL 2025

   Abs PDF Code

   Citation quality is crucial in information-seeking systems, directly influencing trust and the effectiveness of information access. Current evaluation frameworks, both human and automatic, mainly rely on Natural Language Inference (NLI) to assess binary or ternary supportiveness from cited sources, which we argue is a suboptimal proxy for citation evaluation. In this work we introduce CiteEval, a citation evaluation framework driven by principles focusing on fine-grained citation assessment within a broad context, encompassing not only the cited sources but the full retrieval context, user query, and generated text. Guided by the proposed framework, we construct CiteBench, a multi-domain benchmark with high-quality human annotations on citation quality. To enable efficient evaluation, we further develop CiteEval-Auto, a suite of model-based metrics that exhibit strong correlation with human judgments. Experiments across diverse systems demonstrate CiteEval-Auto’s superior ability to capture the multifaceted nature of citations compared to existing metrics, offering a principled and scalable approach to evaluate and improve model-generated citations.

2024

1. LLM - SFT

   Dancing in Chains: Reconciling Instruction Following and Faithfulness in Language Models

   Zhengxuan Wu, Yuhao Zhang*, Peng Qi*, Yumo Xu*, Rujun Han, Yian Zhang, Jifan Chen, Bonan Min, and Zhiheng Huang

   In EMNLP 2024

   Abs PDF Code

   Modern language models (LMs) need to follow human instructions while being faithful; yet, they often fail to achieve both. Here, we provide concrete evidence of a trade-off between instruction following (i.e., follow open-ended instructions) and faithfulness (i.e., ground responses in given context) when training LMs with these objectives. For instance, fine-tuning LLaMA-7B on instruction following datasets renders it less faithful. Conversely, instruction-tuned Vicuna-7B shows degraded performance at following instructions when further optimized on tasks that require contextual grounding. One common remedy is multi-task learning (MTL) with data mixing, yet it remains far from achieving a synergic outcome. We propose a simple yet effective method that relies on Rejection Sampling for Continued Self-instruction Tuning (ReSet), which significantly outperforms vanilla MTL. Surprisingly, we find that less is more, as training ReSet with high-quality, yet substantially smaller data (three-fold less) yields superior results. Our findings offer a better understanding of objective discrepancies in alignment training of LMs.

2. LLM - RAG

   RAG-QA Arena: Evaluating Domain Robustness for Long-form Retrieval Augmented Question Answering

   Rujun Han, Yuhao Zhang, Peng Qi, Yumo Xu, Jenyuan Wang, Lan Liu, William Yang Wang, Bonan Min, and Vittorio Castelli

   In EMNLP 2024

   Abs PDF Code

   Question answering based on retrieval augmented generation (RAG-QA) is an important research topic in NLP and has a wide range of real-world applications. However, most existing datasets for this task are either constructed using a single source corpus or consist of short extractive answers, which fall short of evaluating large language model (LLM) based RAG-QA systems on cross-domain generalization. To address these limitations, we create Long-form RobustQA (LFRQA), a new dataset comprising human-written long-form answers that integrate short extractive answers from multiple documents into a single, coherent narrative, covering 26K queries and large corpora across seven different domains. We further propose RAG-QA Arena by directly comparing model-generated answers against LFRQA’s answers using LLMs as evaluators. We show via extensive experiments that RAG-QA Arena and human judgments on answer quality are highly correlated. Moreover, only 41.3% of the most competitive LLM’s answers are preferred to LFRQA’s answers, demonstrating RAG-QA Arena as a challenging evaluation platform for future research.

3. Summarization

   Fine-Grained Natural Language Inference Based Faithfulness Evaluation for Diverse Summarisation Tasks

   Huajian Zhang, Yumo Xu, and Laura Perez-Beltrachini

   In EACL 2024

   Abs PDF Code

   We study existing approaches to leverage off-the-shelf Natural Language Inference (NLI) models for the evaluation of summary faithfulness and argue that these are sub-optimal due to the granularity level considered for premises and hypotheses. That is, the smaller content unit considered as hypothesis is a sentence and premises are made up of a fixed number of document sentences. We propose a novel approach, namely INFUSE, that uses a variable premise size and simplifies summary sentences into shorter hypotheses. Departing from previous studies which focus on single short document summarisation, we analyse NLI based faithfulness evaluation for diverse summarisation tasks. We introduce DiverSumm, a new benchmark comprising long form summarisation (long documents and summaries) and diverse summarisation tasks (e.g., meeting and multi-document summarisation). In experiments, INFUSE obtains superior performance across the different summarisation tasks.

2023

1. Summarization

   QTSumm: Query-Focused Summarization over Tabular Data

   Yilun Zhao, Zhenting Qi, Linyong Nan, Boyu Mi, Yixin Liu, Weijin Zou, Simeng Han, Xiangru Tang, Yumo Xu, Arman Cohan, and Dragomir Radev

   In EMNLP 2023

   Abs PDF Code

   People primarily consult tables to conduct data analysis or answer specific questions. Text generation systems that can provide accurate table summaries tailored to users’ information needs can facilitate more efficient access to relevant data insights. However, existing table-to-text generation studies primarily focus on converting tabular data into coherent statements, rather than addressing information-seeking purposes. In this paper, we define a new query-focused table summarization task, where text generation models have to perform human-like reasoning and analysis over the given table to generate a tailored summary, and we introduce a new benchmark named QTSumm for this task. QTSumm consists of 5,625 human-annotated query-summary pairs over 2,437 tables on diverse topics. Moreover, we investigate state-of-the-art models (i.e., text generation, table-to-text generation, and large language models) on the QTSumm dataset. Experimental results and manual analysis reveal that our benchmark presents significant challenges in table-to-text generation for future research.

2. Summarization

   Tackling Query-Focused Summarization as A Knowledge-Intensive Task: A Pilot Study

   Zhang Weijia, Svitlana Vakulenko, Thilina Rajapakse, Yumo Xu, and Evangelos Kanoulas

   In Gen-IR@SIGIR 2023

   Abs PDF Code

   Query-focused summarization (QFS) requires generating a summary given a query using a set of relevant documents. However, such relevant documents should be annotated manually and thus are not readily available in realistic scenarios. To address this limitation, we tackle the QFS task as a knowledge-intensive (KI) task without access to any relevant documents. Instead, we assume that these documents are present in a large-scale knowledge corpus and should be retrieved first. To explore this new setting, we build a new dataset (KI-QFS) by adapting existing QFS datasets. In this dataset, answering the query requires document retrieval from a knowledge corpus. We construct three different knowledge corpora, and we further provide relevance annotations to enable retrieval evaluation. Finally, we benchmark the dataset with state-of-the- art QFS models and retrieval-enhanced models. The experimental results demonstrate that QFS models perform significantly worse on KI-QFS compared to the original QFS task, indicating that the knowledge-intensive setting is much more challenging and offers substantial room for improvement. We believe that our investigation will inspire further research into addressing QFS in more realistic scenarios.

3. Summarization

   Abstractive Summarizers are Excellent Extractive Summarizers

   Daniel Varab, and Yumo Xu

   In ACL 2023

   Abs PDF Code

   Extractive and abstractive summarization designs have historically been fragmented, limiting the benefits that often arise from compatible model architectures. In this paper, we explore the potential synergies of modeling extractive summarization with an abstractive summarization system and propose three novel inference algorithms using the sequence-to-sequence architecture. We evaluate them on the CNN & Dailymail dataset and show that recent advancements in abstractive system designs enable abstractive systems to not only compete, but even surpass the performance of extractive systems with custom architectures. To our surprise, abstractive systems achieve this without being exposed to extractive oracle summaries and, therefore, for the first time allow a single model to produce both abstractive and extractive summaries. This evidence questions our fundamental understanding of extractive system design, and the necessity for extractive labels while pathing the way for promising research directions in hybrid models.

4. Summarization

   Text Summarization with Oracle Expectation

   Yumo Xu, and Mirella Lapata

   In ICLR 2023

   Abs PDF Code

   Extractive summarization produces summaries by identifying and concatenating the most important sentences in a document. Since most summarization datasets do not come with gold labels indicating whether document sentences are summary-worthy, different labeling algorithms have been proposed to extrapolate oracle extracts for model training. In this work, we identify two flaws with the widely used greedy labeling approach: it delivers suboptimal and deterministic oracles. To alleviate both issues, we propose a simple yet effective labeling algorithm that creates soft, expectation-based sentence labels. We define a new learning objective for extractive summarization which incorporates learning signals from multiple oracle summaries and prove it is equivalent to estimating the oracle expectation for each document sentence. Without any architectural modifications, the proposed labeling scheme achieves superior performance on a variety of summarization benchmarks across domains and languages, in both supervised and zero-shot settings.

2022

1. Summarization

   Document Summarization with Neural Query Modeling

   Yumo Xu

   2022

   Abs PDF

   Document summarization is a natural language processing task that aims to produce a short summary that concisely delivers the most important information of a document or multiple documents. Over the last few decades, the task has drawn much attention from both academia and industry, as it provides effective tools to manage and access text information. For example, through a newswire summarization engine, users can quickly digest a cluster of news articles by reading a short summary of the topic. Such summaries can, meanwhile, be used by news recommendation and question answering engines. Depending on the users’ role in the summarization process, document summarization falls into two broad categories: generic summarization and query focused summarization (QFS). The former focuses on information intrinsically salient in the input text, while the latter also caters to requests explicitly specified by users. Despite the difference between generic summarization and QFS in their task formulations, we argue that all summaries address queries, even if they are not formulated explicitly. In this thesis, we introduce query modeling in the document summarization context as a critical objective for incorporating observed or latent user intent. We investigate different approaches that explore this theme with deep neural networks. We develop novel systems with neural query modeling for both extractive summarization, where summaries are composed of salient segments (e.g., sentences) from the original document(s), and abstractive summarization, where summaries are made up of words or phrases that do not exist in the input. The recent availability of large-scale datasets has driven the development of neural models that create generic summaries. However, training data in the form of queries, documents, and summaries for QFS is scarce. As most existing research in QFS has employed an extractive approach, we first consider better modeling query-cluster interactions for low-resource extractive QFS. In contrast to previous work with retrieval-style methods for assembling query-relevant summaries, we propose a framework that progressively estimates whether text segments should be included in the summary. Notably, modules of this framework can be independently developed and can leverage training data if available. We present an instantiation of this framework with distant supervision from question answering where various resources exist to identify segments which are likely to answer the query. Experiments on benchmark datasets show that our framework achieves competitive results and is robust across domains. Ideally, summaries should be abstracts, and the hidden costs incurred by annotating QA pairs should be avoided in query modeling. The second part of this thesis focuses on the low-resource challenge in abstractive QFS, and builds an abstractive QFS system which is trained query-free. Concretely, we propose to decompose the task into query modeling and conditional language modeling. For query modeling, we first introduce a unified representation for summaries and queries to exploit training resources in generic summarization, on top of which a weakly supervised model is optimized for evidence estimation. The proposed framework achieves state-of-the-art performance in generating query focused abstracts across existing benchmarks. Finally, the third part of this thesis moves beyond QFS. We provide a unified modeling framework for any kind of summarization, under the assumption that all summaries are a response to a query, which is observed in the case of QFS and latent in the case of generic summarization. We model queries as discrete latent variables over document tokens, and learn representations compatible with observed and unobserved query verbalizations. Requiring no further optimization on downstream summarization tasks, experiments show that our approach outperforms strong comparison systems across benchmarks, query types, document settings, and target domains.

2. Summarization

   Document Summarization with Latent Queries

   Yumo Xu, and Mirella Lapata

   TACL 2022

   Abs PDF Code

   The availability of large-scale datasets has driven the development of neural models that create summaries from single documents, for generic purposes. When using a summarization system, users often have specific intents with various language realizations, which, depending on the information need, can range from a single keyword to a long narrative composed of multiple questions. Existing summarization systems, however, often either fail to support or act robustly on this query focused summarization task. We introduce LaQSum, the first unified text summarization system that learns Latent Queries from documents for abstractive summarization with any existing query forms. Under a deep generative framework, our system jointly optimizes a latent query model and a conditional language model, allowing users to plug-and-play queries of any type at test time. Despite learning from only generic summarization data and requiring no further optimization for downstream summarization tasks, our system robustly outperforms strong comparison systems across summarization benchmarks with different query types, document settings, and target domains.

2021

1. Summarization

   Generating Query Focused Summaries with Query-Free Resources

   Yumo Xu, and Mirella Lapata

   In ACL 2021

   Abs PDF Code

   The availability of large-scale datasets has driven the development of neural sequence-to-sequence models to generate generic summaries, i.e., summaries which do not correspond to any pre-specified queries. However, due to the lack of training data, query focused summarization (QFS) has been studied mainly with extractive methods. In this work, we consider the problem of leveraging only generic summarization resources to build an abstractive QFS system. We propose Marge, a Masked ROUGE Regression framework composed of a novel unified representation for summaries and queries, and a distantly supervised training task for answer evidence estimation. To further utilize generic data for generation, three attributes are incorporated during training and inference to control the shape of the final summary: evidence rank, query guidance, and summary length. Despite learning from minimal supervision, our system achieves state-of-the-art results in the distantly supervised setting across domains and query types.

2020

1. Summarization

   Coarse-to-Fine Query Focused Multi-Document Summarization

   Yumo Xu, and Mirella Lapata

   In EMNLP 2020

   Abs PDF Code

   We consider the problem of better modeling query-cluster interactions to facilitate query focused multi-document summarization. Due to the lack of training data, existing work relies heavily on retrieval-style methods for assembling query relevant summaries. We propose a coarse-to-fine modeling framework which employs progressively more accurate modules for estimating whether text segments are relevant, likely to contain an answer, and central. The modules can be independently developed and leverage training data if available. We present an instantiation of this framework with a trained evidence estimator which relies on distant supervision from question answering (where various resources exist) to identify segments which are likely to answer the query and should be included in the summary. Our framework is robust across domains and query types (i.e., long vs short) and outperforms strong comparison systems on benchmark datasets.

2. Conversation

   Meta Dialogue Policy Learning

   Yumo Xu, Chenguang Zhu, Baolin Peng, and Michael Zeng

   ArXiv 2020

   Abs PDF

   Dialog policy determines the next-step actions for agents and hence is central to a dialogue system. However, when migrated to novel domains with little data, a policy model can fail to adapt due to insufficient interactions with the new environment. We propose Deep Transferable Q-Network (DTQN) to utilize shareable low-level signals between domains, such as dialogue acts and slots. We decompose the state and action representation space into feature subspaces corresponding to these low-level components to facilitate cross-domain knowledge transfer. Furthermore, we embed DTQN in a meta-learning framework and introduce Meta-DTQN with a dual-replay mechanism to enable effective off-policy training and adaptation. In experiments, our model outperforms baseline models in terms of both success rate and dialogue efficiency on the multi-domain dialogue dataset MultiWOZ 2.0.

3. Conversation

   Bootstrapping a Crosslingual Semantic Parser

   Tom Sherborne, Yumo Xu, and Mirella Lapata

   In Findings of EMNLP 2020

   Abs PDF Code

   Recent progress in semantic parsing scarcely considers languages other than English but professional translation can be prohibitively expensive. We adapt a semantic parser trained on a single language, such as English, to new languages and multiple domains with minimal annotation. We query if machine translation is an adequate substitute for training data, and extend this to investigate bootstrapping using joint training with English, paraphrasing, and multilingual pre-trained models. We develop a Transformer-based parser combining paraphrases by ensembling attention over multiple encoders and present new versions of ATIS and Overnight in German and Chinese for evaluation. Experimental results indicate that MT can approximate training data in a new language for accurate parsing when augmented with paraphrasing through multiple MT engines. Considering when MT is inadequate, we also find that using our approach achieves parsing accuracy within 2% of complete translation using only 50% of training data.

2019

1. Summarization

   Weakly Supervised Domain Detection

   Yumo Xu, and Mirella Lapata

   TACL 2019

   Abs PDF Data Code

   In this paper we introduce domain detection as a new natural language processing task. We argue that the ability to detect textual segments that are domain-heavy (i.e., sentences or phrases that are representative of and provide evidence for a given domain) could enhance the robustness and portability of various text classification applications. We propose an encoder-detector framework for domain detection and bootstrap classifiers with multiple instance learning. The model is hierarchically organized and suited to multilabel classification. We demonstrate that despite learning with minimal supervision, our model can be applied to text spans of different granularities, languages, and genres. We also showcase the potential of domain detection for text summarization.

2. Conversation

   Trainable Dynamic Subsampling for End-to-End Speech Recognition

   Shucong Zhang, Erfan Loweimi, Yumo Xu, Peter Bell, and Steve Renals

   In Interspeech 2019

   Abs PDF Code

   Jointly optimised attention-based encoder-decoder models have yielded impressive speech recognition results. The recurrent neural network (RNN) encoder is a key component in such models – it learns the hidden representations of the inputs. However, it is difficult for RNNs to model the long sequences characteristic of speech recognition. To address this, subsampling between stacked recurrent layers of the encoder is commonly employed. This method reduces the length of the input sequence and leads to gains in accuracy. However, static subsampling may both include redundant information and miss relevant information. We propose using a dynamic subsampling RNN (dsRNN) encoder. Unlike a statically subsampled RNN encoder, the dsRNN encoder can learn to skip redundant frames. Furthermore, the skip ratio may vary at different stages of training, thus allowing the encoder to learn the most relevant information for each epoch. Although the dsRNN is unidirectional, it yields lower phone error rates (PERs) than a bidirectional RNN on TIMIT. The dsRNN encoder has a 16.8% PER on the TIMIT test set, a considerable improvement over static subsampling methods used with unidirectional and bidirectional RNN encoders (23.5% and 20.4% PER respectively).

2018

1. Application

   Stock Movement Prediction from Tweets and Historical Prices

   Yumo Xu, and Shay B. Cohen

   In ACL 2018

   Abs PDF Data

   Stock movement prediction is a challenging problem: the market is highly stochastic, and we make temporally-dependent predictions from chaotic data. We treat these three complexities and present a novel deep generative model jointly exploiting text and price signals for this task. Unlike the case with discriminative or topic modeling, our model introduces recurrent, continuous latent variables for a better treatment of stochasticity, and uses neural variational inference to address the intractable posterior inference. We also provide a hybrid objective with temporal auxiliary to flexibly capture predictive dependencies. We demonstrate the state-of-the-art performance of our proposed model on a new stock movement prediction dataset which we collected.