EMNLP 2021 Papers

Performing event and entity coreference resolution across documents vastly increases the number of candidate mentions, making it intractable to do the full n2 pairwise comparisons. Existing approaches simplify by considering coreference only within document clusters, but this fails to handle inter-cluster coreference, common in many applications. As a result cross-document coreference algorithms are rarely applied to downstream tasks. We draw on an insight from discourse coherence theory: potential coreferences are constrained by the reader’s discourse focus. We model the entities/events in a reader’s focus as a neighborhood within a learned latent embedding space which minimizes the distance between mentions and the centroids of their gold coreference clusters. We then use these neighborhoods to sample only hard negatives to train a fine-grained classifier on mention pairs and their local discourse features. Our approach achieves state-of-the-art results for both events and entities on the ECB+, Gun Violence, Football Coreference, and Cross-Domain Cross-Document Coreference corpora. Furthermore, training on multiple corpora improves average performance across all datasets by 17.2 F1 points, leading to a robust coreference resolution model that is now feasible to apply to downstream tasks.

An interpretable system for open-domain reasoning needs to express its reasoning process in a transparent form. Natural language is an attractive representation for this purpose — it is both highly expressive and easy for humans to understand. However, manipulating natural language statements in logically consistent ways is hard: models must cope with variation in how meaning is expressed while remaining precise. In this paper, we describe ParaPattern, a method for building models to generate deductive inferences from diverse natural language inputs without direct human supervision. We train BART-based models (Lewis et al., 2020) to generate the result of applying a particular logical operation to one or more premise statements. Crucially, we develop a largely automated pipeline for constructing suitable training examples from Wikipedia. We evaluate our models using out-of-domain sentence compositions from the QASC (Khot et al., 2020) and EntailmentBank (Dalvi et al., 2021) datasets as well as targeted perturbation sets. Our results show that our models are substantially more accurate and flexible than baseline systems. ParaPattern achieves 85% validity on examples of the ‘substitution’ operation from EntailmentBank without the use of any in-domain training data, matching the performance of a model fine-tuned for EntailmentBank. The full source code for our method is publicly available.

Locating and fixing bugs is a time-consuming task. Most neural machine translation (NMT) based approaches for automatically bug fixing lack generality and do not make full use of the rich information in the source code. In NMT-based bug fixing, we find some predicted code identical to the input buggy code (called unchanged fix) in NMT-based approaches due to high similarity between buggy and fixed code (e.g., the difference may only appear in one particular line). Obviously, unchanged fix is not the correct fix because it is the same as the buggy code that needs to be fixed. Based on these, we propose an intuitive yet effective general framework (called Fix-Filter-Fix or Fˆ3) for bug fixing. Fˆ3 connects models with our filter mechanism to filter out the last model’s unchanged fix to the next. We propose an Fˆ3 theory that can quantitatively and accurately calculate the Fˆ3 lifting effect. To evaluate, we implement the Seq2Seq Transformer (ST) and the AST2Seq Transformer (AT) to form some basic Fˆ3 instances, called Fˆ3_ST+AT and Fˆ3_AT+ST. Comparing them with single model approaches and many model connection baselines across four datasets validates the effectiveness and generality of Fˆ3 and corroborates our findings and methodology.

Finnish is a language with multiple dialects that not only differ from each other in terms of accent (pronunciation) but also in terms of morphological forms and lexical choice. We present the first approach to automatically detect the dialect of a speaker based on a dialect transcript and transcript with audio recording in a dataset consisting of 23 different dialects. Our results show that the best accuracy is received by combining both of the modalities, as text only reaches to an overall accuracy of 57%, where as text and audio reach to 85%. Our code, models and data have been released openly on Github and Zenodo.

Transformers have outperformed recurrent neural networks (RNNs) in natural language generation. But this comes with a signifi- cant computational cost, as the attention mechanism’s complexity scales quadratically with sequence length. Efficient transformer variants have received increasing interest in recent works. Among them, a linear-complexity recurrent variant has proven well suited for autoregressive generation. It approximates the softmax attention with randomized or heuristic feature maps, but can be difficult to train and may yield suboptimal accuracy. This work aims to convert a pretrained transformer into its efficient recurrent counterpart, improving efficiency while maintaining accuracy. Specifically, we propose a swap-then-finetune procedure: in an off-the-shelf pretrained transformer, we replace the softmax attention with its linear-complexity recurrent alternative and then finetune. With a learned feature map, our approach provides an improved tradeoff between efficiency and accuracy over the standard transformer and other recurrent variants. We also show that the finetuning process has lower training cost relative to training these recurrent variants from scratch. As many models for natural language tasks are increasingly dependent on large-scale pretrained transformers, this work presents a viable approach to improving inference efficiency without repeating the expensive pretraining process.

Analyzing microblogs where we post what we experience enables us to perform various applications such as social-trend analysis and entity recommendation. To track emerging trends in a variety of areas, we want to categorize information on emerging entities (e.g., Avatar 2) in microblog posts according to their types (e.g., Film). We thus introduce a new entity typing task that assigns a fine-grained type to each emerging entity when a burst of posts containing that entity is first observed in a microblog. The challenge is to perform typing from noisy microblog posts without relying on prior knowledge of the target entity. To tackle this task, we build large-scale Twitter datasets for English and Japanese using time-sensitive distant supervision. We then propose a modular neural typing model that encodes not only the entity and its contexts but also meta information in multiple posts. To type ‘homographic’ emerging entities (e.g., ‘Go’ means an emerging programming language and a classic board game), which contexts are noisy, we devise a context selector that finds related contexts of the target entity. Experiments on the Twitter datasets confirm the effectiveness of our typing model and the context selector.

Temporal language grounding (TLG) aims to localize a video segment in an untrimmed video based on a natural language description. To alleviate the expensive cost of manual annotations for temporal boundary labels,we are dedicated to the weakly supervised setting, where only video-level descriptions are provided for training. Most of the existing weakly supervised methods generate a candidate segment set and learn cross-modal alignment through a MIL-based framework. However, the temporal structure of the video as well as the complicated semantics in the sentence are lost during the learning. In this work, we propose a novel candidate-free framework: Fine-grained Semantic Alignment Network (FSAN), for weakly supervised TLG. Instead of view the sentence and candidate moments as a whole, FSAN learns token-by-clip cross-modal semantic alignment by an iterative cross-modal interaction module, generates a fine-grained cross-modal semantic alignment map, and performs grounding directly on top of the map. Extensive experiments are conducted on two widely-used benchmarks: ActivityNet-Captions, and DiDeMo, where our FSAN achieves state-of-the-art performance.

Factual inconsistencies existed in the output of abstractive summarization models with original documents are frequently presented. Fact consistency assessment requires the reasoning capability to find subtle clues to identify whether a model-generated summary is consistent with the original document. This paper proposes a fine-grained two-stage Fact Consistency assessment framework for Summarization models (SumFC). Given a document and a summary sentence, in the first stage, SumFC selects the top-K most relevant sentences with the summary sentence from the document. In the second stage, the model performs fine-grained consistency reasoning at the sentence level, and then aggregates all sentences’ consistency scores to obtain the final assessment result. We get the training data pairs by data synthesis and adopt contrastive loss of data pairs to help the model identify subtle cues. Experiment results show that SumFC has made a significant improvement over the previous state-of-the-art methods. Our experiments also indicate that SumFC distinguishes detailed differences better.

Existing work on Fine-grained Entity Typing (FET) typically trains automatic models on the datasets obtained by using Knowledge Bases (KB) as distant supervision. However, the reliance on KB means this training setting can be hampered by the lack of or the incompleteness of the KB. To alleviate this limitation, we propose a novel setting for training FET models: FET without accessing any knowledge base. Under this setting, we propose a two-step framework to train FET models. In the first step, we automatically create pseudo data with fine-grained labels from a large unlabeled dataset. Then a neural network model is trained based on the pseudo data, either in an unsupervised way or using self-training under the weak guidance from a coarse-grained Named Entity Recognition (NER) model. Experimental results show that our method achieves competitive performance with respect to the models trained on the original KB-supervised datasets.

Conventional entity typing approaches are based on independent classification paradigms, which make them difficult to recognize inter-dependent, long-tailed and fine-grained entity types. In this paper, we argue that the implicitly entailed extrinsic and intrinsic dependencies between labels can provide critical knowledge to tackle the above challenges. To this end, we propose Label Reasoning Network(LRN), which sequentially reasons fine-grained entity labels by discovering and exploiting label dependencies knowledge entailed in the data. Specifically, LRN utilizes an auto-regressive network to conduct deductive reasoning and a bipartite attribute graph to conduct inductive reasoning between labels, which can effectively model, learn and reason complex label dependencies in a sequence-to-set, end-to-end manner. Experiments show that LRN achieves the state-of-the-art performance on standard ultra fine-grained entity typing benchmarks, and can also resolve the long tail label problem effectively.

Modern semantic parsers suffer from two principal limitations. First, training requires expensive collection of utterance-program pairs. Second, semantic parsers fail to generalize at test time to new compositions/structures that have not been observed during training. Recent research has shown that automatic generation of synthetic utterance-program pairs can alleviate the first problem, but its potential for the second has thus far been under-explored. In this work, we investigate automatic generation of synthetic utterance-program pairs for improving compositional generalization in semantic parsing. Given a small training set of annotated examples and an “infinite” pool of synthetic examples, we select a subset of synthetic examples that are structurally-diverse and use them to improve compositional generalization. We evaluate our approach on a new split of the schema2QA dataset, and show that it leads to dramatic improvements in compositional generalization as well as moderate improvements in the traditional i.i.d setup. Moreover, structurally-diverse sampling achieves these improvements with as few as 5K examples, compared to 1M examples when sampling uniformly at random – a 200x improvement in data efficiency.

Human evaluation for summarization tasks is reliable but brings in issues of reproducibility and high costs. Automatic metrics are cheap and reproducible but sometimes poorly correlated with human judgment. In this work, we propose flexible semiautomatic to automatic summary evaluation metrics, following the Pyramid human evaluation method. Semi-automatic Lite2Pyramid retains the reusable human-labeled Summary Content Units (SCUs) for reference(s) but replaces the manual work of judging SCUs’ presence in system summaries with a natural language inference (NLI) model. Fully automatic Lite3Pyramid further substitutes SCUs with automatically extracted Semantic Triplet Units (STUs) via a semantic role labeling (SRL) model. Finally, we propose in-between metrics, Lite2.xPyramid, where we use a simple regressor to predict how well the STUs can simulate SCUs and retain SCUs that are more difficult to simulate, which provides a smooth transition and balance between automation and manual evaluation. Comparing to 15 existing metrics, we evaluate human-metric correlations on 3 existing meta-evaluation datasets and our newly collected PyrXSum (with 100/10 XSum examples/systems). It shows that Lite2Pyramid consistently has the best summary-level correlations; Lite3Pyramid works better than or comparable to other automatic metrics; Lite2.xPyramid trades off small correlation drops for larger manual effort reduction, which can reduce costs for future data collection.

The sheer volume of financial statements makes it difficult for humans to access and analyze a business’s financials. Robust numerical reasoning likewise faces unique challenges in this domain. In this work, we focus on answering deep questions over financial data, aiming to automate the analysis of a large corpus of financial documents. In contrast to existing tasks on general domain, the finance domain includes complex numerical reasoning and understanding of heterogeneous representations. To facilitate analytical progress, we propose a new large-scale dataset, FinQA, with Question-Answering pairs over Financial reports, written by financial experts. We also annotate the gold reasoning programs to ensure full explainability. We further introduce baselines and conduct comprehensive experiments in our dataset. The results demonstrate that popular, large, pre-trained models fall far short of expert humans in acquiring finance knowledge and in complex multi-step numerical reasoning on that knowledge. Our dataset – the first of its kind – should therefore enable significant, new community research into complex application domains. The dataset and code are publicly available at https://github.com/czyssrs/FinQA.

We present models which complete missing text given transliterations of ancient Mesopotamian documents, originally written on cuneiform clay tablets (2500 BCE - 100 CE). Due to the tablets’ deterioration, scholars often rely on contextual cues to manually fill in missing parts in the text in a subjective and time-consuming process. We identify that this challenge can be formulated as a masked language modelling task, used mostly as a pretraining objective for contextualized language models. Following, we develop several architectures focusing on the Akkadian language, the lingua franca of the time. We find that despite data scarcity (1M tokens) we can achieve state of the art performance on missing tokens prediction (89% hit@5) using a greedy decoding scheme and pretraining on data from other languages and different time periods. Finally, we conduct human evaluations showing the applicability of our models in assisting experts to transcribe texts in extinct languages.

With the emergence of the COVID-19 pandemic, the political and the medical aspects of disinformation merged as the problem got elevated to a whole new level to become the first global infodemic. Fighting this infodemic has been declared one of the most important focus areas of the World Health Organization, with dangers ranging from promoting fake cures, rumors, and conspiracy theories to spreading xenophobia and panic. Addressing the issue requires solving a number of challenging problems such as identifying messages containing claims, determining their check-worthiness and factuality, and their potential to do harm as well as the nature of that harm, to mention just a few. To address this gap, we release a large dataset of 16K manually annotated tweets for fine-grained disinformation analysis that (i) focuses on COVID-19, (ii) combines the perspectives and the interests of journalists, fact-checkers, social media platforms, policy makers, and society, and (iii) covers Arabic, Bulgarian, Dutch, and English. Finally, we show strong evaluation results using pretrained Transformers, thus confirming the practical utility of the dataset in monolingual vs. multilingual, and single task vs. multitask settings.

Automatic hate speech detection is hampered by the scarcity of labeled datasetd, leading to poor generalization. We employ pretrained language models (LMs) to alleviate this data bottleneck. We utilize the GPT LM for generating large amounts of synthetic hate speech sequences from available labeled examples, and leverage the generated data in fine-tuning large pretrained LMs on hate detection. An empirical study using the models of BERT, RoBERTa and ALBERT, shows that this approach improves generalization significantly and consistently within and across data distributions. In fact, we find that generating relevant labeled hate speech sequences is preferable to using out-of-domain, and sometimes also within-domain, human-labeled examples.

Natural language (NL) explanations of model predictions are gaining popularity as a means to understand and verify decisions made by large black-box pre-trained models, for tasks such as Question Answering (QA) and Fact Verification. Recently, pre-trained sequence to sequence (seq2seq) models have proven to be very effective in jointly making predictions, as well as generating NL explanations. However, these models have many shortcomings; they can fabricate explanations even for incorrect predictions, they are difficult to adapt to long input documents, and their training requires a large amount of labeled data. In this paper, we develop FiD-Ex, which addresses these shortcomings for seq2seq models by: 1) introducing sentence markers to eliminate explanation fabrication by encouraging extractive generation, 2) using the fusion-in-decoder architecture to handle long input contexts, and 3) intermediate fine-tuning on re-structured open domain QA datasets to improve few-shot performance. FiD-Ex significantly improves over prior work in terms of explanation metrics and task accuracy on five tasks from the ERASER explainability benchmark in both fully supervised and few-shot settings.

The task of learning from only a few examples (called a few-shot setting) is of key importance and relevance to a real-world setting. For question answering (QA), the current state-of-the-art pre-trained models typically need fine-tuning on tens of thousands of examples to obtain good results. Their performance degrades significantly in a few-shot setting (< 100 examples). To address this, we propose a simple fine-tuning framework that leverages pre-trained text-to-text models and is directly aligned with their pre-training framework. Specifically, we construct the input as a concatenation of the question, a mask token representing the answer span and a context. Given this input, the model is fine-tuned using the same objective as that of its pre-training objective. Through experimental studies on various few-shot configurations, we show that this formulation leads to significant gains on multiple QA benchmarks (an absolute gain of 34.2 F1 points on average when there are only 16 training examples). The gains extend further when used with larger models (Eg:- 72.3 F1 on SQuAD using BART-large with only 32 examples) and translate well to a multilingual setting . On the multilingual TydiQA benchmark, our model outperforms the XLM-Roberta-large by an absolute margin of upto 40 F1 points and an average of 33 F1 points in a few-shot setting (<= 64 training examples). We conduct detailed ablation studies to analyze factors contributing to these gains.

Providing pretrained language models with simple task descriptions in natural language enables them to solve some tasks in a fully unsupervised fashion. Moreover, when combined with regular learning from examples, this idea yields impressive few-shot results for a wide range of text classification tasks. It is also a promising direction to improve data efficiency in generative settings, but there are several challenges to using a combination of task descriptions and example-based learning for text generation. In particular, it is crucial to find task descriptions that are easy to understand for the pretrained model and to ensure that it actually makes good use of them; furthermore, effective measures against overfitting have to be implemented. In this paper, we show how these challenges can be tackled: We introduce GenPET, a method for text generation that is based on pattern-exploiting training, a recent approach for combining textual instructions with supervised learning that only works for classification tasks. On several summarization and headline generation datasets, GenPET gives consistent improvements over strong baselines in few-shot settings.