Previous research in multi-document news summarization has typically concentrated on collating information that all sources agree upon. However, the summarization of diverse information dispersed across multiple articles about an event remains underexplored. In this paper, we propose a new task of summarizing diverse information encountered in multiple news articles encompassing the same event. To facilitate this task, we outlined a data collection schema for identifying diverse information and curated a dataset named DiverseSumm. The dataset includes 245 news stories, with each story comprising 10 news articles and paired with a human-validated reference. Next, to enable consistent automatic evaluation, we conducted a comprehensive analysis to pinpoint the position and verbosity biases when utilizing Large Language Model (LLM)-based metrics for evaluating the coverage and faithfulness of summaries. Through correlation analyses, we outline the best practices for effectively using automatic LLM-based metrics on the DiverseSumm dataset. Finally, we study how LLMs summarize multiple news articles by analyzing which type of diverse information LLMs are capable of identifying. Our analyses suggest that despite the extraordinary capabilities of LLMs in single-document summarization, the proposed task remains a complex challenge for them mainly due to their limited coverage, with GPT-4 only able to cover under 40% of the diverse information on average.

Hallucination is a known issue for neural abstractive summarization models. Recent work suggests that the degree of hallucination may depend on factual errors in the training data. In this work, we propose a new method called Contrastive Parameter Ensembling (CaPE) to use training data more effectively, utilizing variations in noise in training samples to reduce hallucination. Starting with a base model fine-tuned on an entire dataset, we additionally train expert and anti-expert models on clean and noisy subsets of the data, respectively. We then adjust the parameters of the base model by adding (subtracting) the parameters of the expert (anti-expert), advancing the recent work on additive parameter ensembling approaches. Trained on a much smaller data subset, expert and anti-expert models only fractionally (<14%) increases the total training time. Further, CaPE uses parameter ensembling and does not increase the inference time. Experimental results show that CaPE improves performance across different automatic factual metrics and human evaluation, with a maximum improvement of 16.69% and 15.38% on summary-level dependency-arc entailment accuracy for the XSUM and CNN/DM datasets. The CaPE model performs comparably to the base model on metrics of informativeness such as ROUGE.

Ideal summarization models should generalize to novel summary-worthy content without remembering reference training summaries by rote. However, a single average performance score on the entire test set is inadequate in determining such model competencies. We propose a fine-grained evaluation protocol by partitioning a test set based on the lexical similarity of reference test summaries with training summaries. We observe up to a 5x (1.2x) difference in ROUGE-2 (entity recall) scores between the subsets with the lowest and highest similarity. Next, we show that such training repetitions also make a model vulnerable to rote learning, reproducing data artifacts such as factual errors, especially when reference test summaries are lexically close to training summaries. Consequently, we propose to limit lexical repetitions in training summaries during both supervised fine-tuning and likelihood calibration stages to improve the performance on novel test cases while retaining average performance. Our automatic and human evaluations on novel test subsets and recent news articles show that limiting lexical repetitions in training summaries can prevent rote learning and improve generalization.

Document-level text simplification often deletes some sentences besides performing lexical, grammatical or structural simplification to reduce text complexity. In this work, we focus on sentence deletions for text simplification and use a news genre-specific functional discourse structure, which categorizes sentences based on their contents and their function roles in telling a news story, for predicting sentence deletion. We incorporate sentence categories into a neural net model in two ways for predicting sentence deletions, either as additional features or by jointly predicting sentence deletions and sentence categories. Experimental results using human-annotated data show that incorporating the functional structure improves the recall of sentence deletion prediction by 6.5% and 10.7% respectively using the two methods, and improves the overall F1-score by 3.6% and 4.3% respectively.

Pre-trained language models (PLMs) have been shown effective for zero-shot (0shot) text classification. 0shot models based on natural language inference (NLI) and next sentence prediction (NSP) employ cross-encoder architecture and infer by making a forward pass through the model for each label-text pair separately. This increases the computational cost to make inferences linearly in the number of labels. In this work, we improve the efficiency of such cross-encoder-based 0shot models by restricting the number of likely labels using another fast base classifier-based conformal predictor (CP) calibrated on samples labeled by the 0shot model. Since a CP generates prediction sets with coverage guarantees, it reduces the number of target labels without excluding the most probable label based on the 0shot model. We experiment with three intent and two topic classification datasets. With a suitable CP for each dataset, we reduce the average inference time for NLI- and NSP-based models by 25.6% and 22.2% respectively, without dropping performance below the predefined error rate of 1%.

State-of-the-art summarization models still struggle to be factually consistent with the input text. A model-agnostic way to address this problem is post-editing the generated summaries. However, existing approaches typically fail to remove entity errors if a suitable input entity replacement is not available or may insert erroneous content. In our work, we focus on removing extrinsic entity errors, or entities not in the source, to improve consistency while retaining the summary’s essential information and form. We propose to use sentence-compression data to train the post-editing model to take a summary with extrinsic entity errors marked with special tokens and output a compressed, well-formed summary with those errors removed. We show that this model improves factual consistency while maintaining ROUGE, improving entity precision by up to 30% on XSum, and that this model can be applied on top of another post-editor, improving entity precision by up to a total of 38%. We perform an extensive comparison of post-editing approaches that demonstrate trade-offs between factual consistency, informativeness, and grammaticality, and we analyze settings where post-editors show the largest improvements.

We propose to leverage news discourse profiling to model document-level temporal structures for building temporal dependency graphs. Our key observation is that the functional roles of sentences used for profiling news discourse signify different time frames relevant to a news story and can, therefore, help to recover the global temporal structure of a document. Our analyses and experiments with the widely used knowledge distillation technique show that discourse profiling effectively identifies distant inter-sentence event and (or) time expression pairs that are temporally related and otherwise difficult to locate.

We propose to leverage lexical paraphrases and high precision rules informed by news discourse structure to automatically collect coreferential and non-coreferential event pairs from unlabeled English news articles. We perform both manual validation and empirical evaluation on multiple evaluation datasets with different event domains and text genres to assess the quality of our acquired event pairs. We found that a model trained on our acquired event pairs performs comparably as the supervised model when applied to new data out of the training data domains. Further, augmenting human-annotated data with the acquired event pairs provides empirical performance gains on both in-domain and out-of-domain evaluation datasets.

We present an actor-critic framework to induce subtopical structures in a news article for news discourse profiling. The model uses multiple critics that act according to known subtopic structures while the actor aims to outperform them. The content structures constitute sentences that represent latent subtopic boundaries. Then, we introduce a hierarchical neural network that uses the identified subtopic boundary sentences to model multi-level interaction between sentences, subtopics, and the document. Experimental results and analyses on the NewsDiscourse corpus show that the actor model learns to effectively segment a document into subtopics and improves the performance of the hierarchical model on the news discourse profiling task.

Targeted evaluations have found that machine translation systems often output incorrect gender in translations, even when the gender is clear from context. Furthermore, these incorrectly gendered translations have the potential to reflect or amplify social biases. We propose gender-filtered self-training (GFST) to improve gender translation accuracy on unambiguously gendered inputs. Our GFST approach uses a source monolingual corpus and an initial model to generate gender-specific pseudo-parallel corpora which are then filtered and added to the training data. We evaluate GFST on translation from English into five languages, finding that it improves gender accuracy without damaging generic quality. We also show the viability of GFST on several experimental settings, including re-training from scratch, fine-tuning, controlling the gender balance of the data, forward translation, and back-translation.

Most supervised word sense disambiguation (WSD) systems build word-specific classifiers by leveraging labeled data. However, when using word-specific classifiers, the sparseness of annotations leads to inferior sense disambiguation performance on less frequently seen words. To combat data sparsity, we propose to learn a single model that derives sense representations and meanwhile enforces congruence between a word instance and its right sense by using both sense-annotated data and lexical resources. The model is shared across words that allows utilizing sense correlations across words, and therefore helps to transfer common disambiguation rules from annotation-rich words to annotation-lean words. Empirical evaluation on benchmark datasets shows that the proposed shared model outperforms the equivalent classifier-based models by 1.7%, 2.5% and 3.8% in F1-score when using GloVe, ELMo and BERT word embeddings respectively.

Understanding discourse structures of news articles is vital to effectively contextualize the occurrence of a news event. To enable computational modeling of news structures, we apply an existing theory of functional discourse structure for news articles that revolves around the main event and create a human-annotated corpus of 802 documents spanning over four domains and three media sources. Next, we propose several document-level neural-network models to automatically construct news content structures. Finally, we demonstrate that incorporating system predicted news structures yields new state-of-the-art performance for event coreference resolution. The news documents we annotated are openly available and the annotations are publicly released for future research.

A typical conversation comprises of multiple turns between participants where they go back and forth between different topics. At each user turn, dialogue state tracking (DST) aims to estimate user’s goal by processing the current utterance. However, in many turns, users implicitly refer to the previous goal, necessitating the use of relevant dialogue history. Nonetheless, distinguishing relevant history is challenging and a popular method of using dialogue recency for that is inefficient. We, therefore, propose a novel framework for DST that identifies relevant historical context by referring to the past utterances where a particular slot-value changes and uses that together with weighted system utterance to identify the relevant context. Specifically, we use the current user utterance and the most recent system utterance to determine the relevance of a system utterance. Empirical analyses show that our method improves joint goal accuracy by 2.75% and 2.36% on WoZ 2.0 and Multi-WoZ restaurant domain datasets respectively over the previous state-of-the-art GLAD model.

We aim to comprehensively identify all the event causal relations in a document, both within a sentence and across sentences, which is important for reconstructing pivotal event structures. The challenges we identified are two: 1) event causal relations are sparse among all possible event pairs in a document, in addition, 2) few causal relations are explicitly stated. Both challenges are especially true for identifying causal relations between events across sentences. To address these challenges, we model rich aspects of document-level causal structures for achieving comprehensive causal relation identification. The causal structures include heavy involvements of document-level main events in causal relations as well as several types of fine-grained constraints that capture implications from certain sentential syntactic relations and discourse relations as well as interactions between event causal relations and event coreference relations. Our experimental results show that modeling the global and fine-grained aspects of causal structures using Integer Linear Programming (ILP) greatly improves the performance of causal relation identification, especially in identifying cross-sentence causal relations.

The increasing prevalence of political bias in news media calls for greater public awareness of it, as well as robust methods for its detection. While prior work in NLP has primarily focused on the lexical bias captured by linguistic attributes such as word choice and syntax, other types of bias stem from the actual content selected for inclusion in the text. In this work, we investigate the effects of informational bias: factual content that can nevertheless be deployed to sway reader opinion. We first produce a new dataset, BASIL, of 300 news articles annotated with 1,727 bias spans and find evidence that informational bias appears in news articles more frequently than lexical bias. We further study our annotations to observe how informational bias surfaces in news articles by different media outlets. Lastly, a baseline model for informational bias prediction is presented by fine-tuning BERT on our labeled data, indicating the challenges of the task and future directions.

Identifying the most dominant and central event of a document, which governs and connects other foreground and background events in the document, is useful for many applications, such as text summarization, storyline generation and text segmentation. We observed that the central event of a document usually has many coreferential event mentions that are scattered throughout the document for enabling a smooth transition of subtopics. Our empirical experiments, using gold event coreference relations, have shown that the central event of a document can be well identified by mining properties of event coreference chains. But the performance drops when switching to system predicted event coreference relations. In addition, we found that the central event can be more accurately identified by further considering the number of sub-events as well as the realis status of an event.

This paper proposes a novel approach for event coreference resolution that models correlations between event coreference chains and document topical structures through an Integer Linear Programming formulation. We explicitly model correlations between the main event chains of a document with topic transition sentences, inter-coreference chain correlations, event mention distributional characteristics and sub-event structure, and use them with scores obtained from a local coreference relation classifier for jointly resolving multiple event chains in a document. Our experiments across KBP 2016 and 2017 datasets suggest that each of the structures contribute to improving event coreference resolution performance.

We present a sequential model for temporal relation classification between intra-sentence events. The key observation is that the overall syntactic structure and compositional meanings of the multi-word context between events are important for distinguishing among fine-grained temporal relations. Specifically, our approach first extracts a sequence of context words that indicates the temporal relation between two events, which well align with the dependency path between two event mentions. The context word sequence, together with a parts-of-speech tag sequence and a dependency relation sequence that are generated corresponding to the word sequence, are then provided as input to bidirectional recurrent neural network (LSTM) models. The neural nets learn compositional syntactic and semantic representations of contexts surrounding the two events and predict the temporal relation between them. Evaluation of the proposed approach on TimeBank corpus shows that sequential modeling is capable of accurately recognizing temporal relations between events, which outperforms a neural net model using various discrete features as input that imitates previous feature based models.

We introduce a novel iterative approach for event coreference resolution that gradually builds event clusters by exploiting inter-dependencies among event mentions within the same chain as well as across event chains. Among event mentions in the same chain, we distinguish within- and cross-document event coreference links by using two distinct pairwise classifiers, trained separately to capture differences in feature distributions of within- and cross-document event clusters. Our event coreference approach alternates between WD and CD clustering and combines arguments from both event clusters after every merge, continuing till no more merge can be made. And then it performs further merging between event chains that are both closely related to a set of other chains of events. Experiments on the ECB+ corpus show that our model outperforms state-of-the-art methods in joint task of WD and CD event coreference resolution.