Evaluation is a cornerstone topic in NLP. However, many criticisms have been raised about the community’s evaluation practices, including a lack of human-centered considerations about people’s needs for language technologies and their actual impact on people. This “evaluation crisis” is exacerbated by the recent development of large generative models with diverse and uncertain capabilities. This tutorial aims to inspire more human-centered evaluation in NLP by introducing perspectives and methodologies from human-computer interaction (HCI), a field concerned primarily with the design and evaluation of technologies. The tutorial will start with an overview of current NLP evaluation practices and their limitations, then introduce the “toolbox of evaluation methods” from HCI with varying considerations such as what to evaluate for, how generalizable the results are to the real-world contexts, and pragmatic costs to conduct the evaluation. The tutorial will also encourage reflection on how these HCI perspectives and methodologies can complement NLP evaluation through Q&A discussions and a hands-on exercise.

It is increasingly common to evaluate the same coreference resolution (CR) model on multiple datasets. Do these multi-dataset evaluations allow us to draw meaningful conclusions about model generalization? Or, do they rather reflect the idiosyncrasies of a particular experimental setup (e.g., the specific datasets used)? To study this, we view evaluation through the lens of measurement modeling, a framework commonly used in the social sciences for analyzing the validity of measurements. By taking this perspective, we show how multi-dataset evaluations risk conflating different factors concerning what, precisely, is being measured. This in turn makes it difficult to draw more generalizable conclusions from these evaluations. For instance, we show that across seven datasets, measurements intended to reflect CR model generalization are often correlated with differences in both how coreference is defined and how it is operationalized; this limits our ability to draw conclusions regarding the ability of CR models to generalize across any singular dimension. We believe the measurement modeling framework provides the needed vocabulary for discussing challenges surrounding what is actually being measured by CR evaluations.

Recent progress in large language models (LLMs) has led to their widespread adoption in various domains. However, these advancements have also introduced additional safety risks and raised concerns regarding their detrimental impact on already marginalized populations.Despite growing mitigation efforts to develop safety safeguards, such as supervised safety-oriented fine-tuning and leveraging safe reinforcement learning from human feedback, multiple concerns regarding the safety and ingrained biases in these models remain. Furthermore, previous work has demonstrated that models optimized for safety often display exaggerated safety behaviors, such as a tendency to refrain from responding to certain requests as a precautionary measure. As such, a clear trade-off between the helpfulness and safety of these models has been documented in the literature. In this paper, we further investigate the effectiveness of safety measures by evaluating models on already mitigated biases. Using the case of Llama 2 as an example, we illustrate how LLMs’ safety responses can still encode harmful assumptions. To do so, we create a set of non-toxic prompts, which we then use to evaluate Llama models. Through our new taxonomy of LLMs responses to users, we observe that the safety/helpfulness trade-offs are more pronounced for certain demographic groups which can lead to different kinds of harms such as quality-of-service harms for marginalized populations.

Current end-to-end coreference resolution models combine detection of singleton mentions and antecedent linking into a single step. In contrast, singleton detection was often treated as a separate step in the pre-neural era. In this work, we show that separately parameterizing these two sub-tasks also benefits end-to-end neural coreference systems. Specifically, we add a singleton detector to the coarse-to-fine (C2F) coreference model, and design an anaphoricity-aware span embedding and singleton detection loss. Our method significantly improves model performance on OntoNotes and four additional datasets.

Assessing the quality of summarizers poses significant challenges—gold summaries are hard to obtain and their suitability depends on the use context of the summarization system. Who is the user of the system, and what do they intend to do with the summary? In response, we propose a novel task-oriented evaluation approach that assesses summarizers based on their capacity to produce summaries while preserving task outcomes. We theoretically establish both a lower and upper bound on the expected error rate of these tasks, which depends on the mutual information between source texts and generated summaries. We introduce COSMIC, a practical implementation of this metric, and demonstrate its strong correlation with human judgment-based metrics, as well as its effectiveness in predicting downstream task performance. Comparative analyses against established metrics like BERTScore and ROUGE highlight the competitive performance of COSMIC.

Scientific peer review is essential for the quality of academic publications. However, the increasing number of paper submissions to conferences has strained the reviewing process. This surge poses a burden on area chairs who have to carefully read an ever-growing volume of reviews and discern each reviewer’s main arguments as part of their decision process. In this paper, we introduce , a summarization method designed to offer a concise yet comprehensive overview of scholarly reviews. Unlike traditional consensus-based methods, extracts both common and unique opinions from the reviews. We introduce novel uniqueness scores based on the Rational Speech Act framework to identify relevant sentences in the reviews. Our method aims to provide a pragmatic glimpse into all reviews, offering a balanced perspective on their opinions. Our experimental results with both automatic metrics and human evaluation show that generates more discriminative summaries than baseline methods in terms of human evaluation while achieving comparable performance with these methods in terms of automatic metrics.

Benchmarking is seen as critical to assessing progress in NLP. However, creating a benchmark involves many design decisions (e.g., which datasets to include, which metrics to use) that often rely on tacit, untested assumptions about what the benchmark is intended to measure or is actually measuring. There is currently no principled way of analyzing these decisions and how they impact the validity of the benchmark’s measurements. To address this gap, we draw on evidence-centered design in educational assessments and propose Evidence-Centered Benchmark Design (ECBD), a framework which formalizes the benchmark design process into five modules. ECBD specifies the role each module plays in helping practitioners collect evidence about capabilities of interest. Specifically, each module requires benchmark designers to describe, justify, and support benchmark design choices—e.g., clearly specifying the capabilities the benchmark aims to measure or how evidence about those capabilities is collected from model responses. To demonstrate the use of ECBD, we conduct case studies with three benchmarks: BoolQ, SuperGLUE, and HELM. Our analysis reveals common trends in benchmark design and documentation that could threaten the validity of benchmarks’ measurements.

All state-of-the-art coreference resolution (CR) models involve finetuning a pretrained language model. Whether the superior performance of one CR model over another is due to the choice of language model or other factors, such as the task-specific architecture, is difficult or impossible to determine due to lack of a standardized experimental setup. To resolve this ambiguity, we systematically evaluate five CR models and control for certain design decisions including the pretrained language model used by each. When controlling for language model size, encoder-based CR models outperform more recent decoder-based models in terms of both accuracy and inference speed. Surprisingly, among encoder-based CR models, more recent models are not always more accurate, and the oldest CR model that we test generalizes the best to out-of-domain textual genres. We conclude that controlling for the choice of language model reduces most, but not all, of the increase in F1 score reported in the past five years.

Many state-of-the-art natural language understanding (NLU) models are based on pretrained neural language models. These models often make inferences using information from multiple sources. An important class of such inferences are those that require both background knowledge, presumably contained in a model’s pretrained parameters, and instance-specific information that is supplied at inference time. However, the integration and reasoning abilities of NLU models in the presence of multiple knowledge sources have been largely understudied. In this work, we propose a test suite of coreference resolution subtasks that require reasoning over multiple facts. These subtasks differ in terms of which knowledge sources contain the relevant facts. We also introduce subtasks where knowledge is present only at inference time using fictional knowledge. We evaluate state-of-the-art coreference resolution models on our dataset. Our results indicate that several models struggle to reason on-the-fly over knowledge observed both at pretrain time and at inference time. However, with task-specific training, a subset of models demonstrates the ability to integrate certain knowledge types from multiple sources. Still, even the best performing models seem to have difficulties with reliably integrating knowledge presented only at inference time.

We present Varta, a large-scale multilingual dataset for headline generation in Indic languages. This dataset includes more than 41 million pairs of headlines and articles in 14 different Indic languages (and English), which come from a variety of high-quality news sources. To the best of our knowledge, this is the largest collection of curated news articles for Indic languages currently available. We use the collected data in a series of experiments to answer important questions related to Indic NLP and multilinguality research in general. We show that the dataset is challenging even for state-of-the-art abstractive models and that they perform only slightly better than extractive baselines. Owing to its size, we also show that the dataset can be used to pre-train strong language models that outperform competitive baselines in both NLU and NLG benchmarks.

AI and NLP publication venues have increasingly encouraged researchers to reflect on possible ethical considerations, adverse impacts, and other responsible AI issues their work might engender. However, for specific NLP tasks our understanding of how prevalent such issues are, or when and why these issues are likely to arise, remains limited. Focusing on text summarization—a common NLP task largely overlooked by the responsible AI community—we examine research and reporting practices in the current literature. We conduct a multi-round qualitative analysis of 333 summarization papers from the ACL Anthology published between 2020–2022. We focus on how, which, and when responsible AI issues are covered, which relevant stakeholders are considered, and mismatches between stated and realized research goals. We also discuss current evaluation practices and consider how authors discuss the limitations of both prior work and their own work. Overall, we find that relatively few papers engage with possible stakeholders or contexts of use, which limits their consideration of potential downstream adverse impacts or other responsible AI issues. Based on our findings, we make recommendations on concrete practices and research directions.

The potential of using a large language model (LLM) as a knowledge base (KB) has sparked significant interest. To maintain the knowledge acquired by LLMs, we need to ensure that the editing of learned facts respects internal logical constraints, which are known as dependency of knowledge. Existing work on editing LLMs has partially addressed the issue of dependency, when the editing of a fact should apply to its lexical variations without disrupting irrelevant ones. However, they neglect the dependency between a fact and its logical implications. We propose an evaluation protocol with an accompanying question-answering dataset, StandUp, that provides a comprehensive assessment of the editing process considering the above notions of dependency. Our protocol involves setting up a controlled environment in which we edit facts and monitor their impact on LLMs, along with their implications based on If-Then rules. Extensive experiments on StandUp show that existing knowledge editing methods are sensitive to the surface form of knowledge, and that they have limited performance in inferring the implications of edited facts.

Lexical semantic relations (LSRs) characterize meaning relationships between words and play an important role in systematic generalization on lexical inference tasks. Notably, several tasks that require knowledge of hypernymy still pose a challenge for pretrained language models (LMs) such as BERT, underscoring the need to better align their linguistic behavior with our knowledge of LSRs. In this paper, we propose Balaur, a model that addresses this challenge by modeling LSRs directly in the LM’s hidden states throughout pretraining. Motivating our approach is the hypothesis that the internal representations of LMs can provide an interface to their observable linguistic behavior, and that by controlling one we can influence the other. We validate our hypothesis and demonstrate that Balaur generally improves the performance of large transformer-based LMs on a comprehensive set of hypernymy-informed tasks, as well as on the original LM objective. Code and data are made available at https://github.com/mirandrom/balaur

We explore the connection between presupposition, discourse and sarcasm and propose to leverage that connection in a transfer learning scenario with the goal of improving the performance of NLI models on cases involving presupposition. We exploit advances in training transformer-based models that show that pre-finetuning—–i.e., finetuning the model on an additional task or dataset before the actual finetuning phase—–can help these models, in some cases, achieve a higher performance on a given downstream task. Building on those advances and that aforementioned connection, we propose pre-finetuning NLI models on carefully chosen tasks in an attempt to improve their performance on NLI cases involving presupposition. We notice that, indeed, pre-finetuning on those tasks leads to performance improvements. Furthermore, we run several diagnostic tests to understand whether these gains are merely a byproduct of additional training data. The results show that, while additional training data seems to be helping on its own in some cases, the choice of the tasks plays a role in the performance improvements.

Qualitative coding is a content analysis method in which researchers read through a text corpus and assign descriptive labels or qualitative codes to passages. It is an arduous and manual process which human-computer interaction (HCI) studies have shown could greatly benefit from NLP techniques to assist qualitative coders. Yet, previous attempts at leveraging language technologies have set up qualitative coding as a fully automatable classification problem. In this work, we take a more assistive approach by defining the task of qualitative code suggestion (QCS) in which a ranked list of previously assigned qualitative codes is suggested from an identified passage. In addition to being user-motivated, QCS integrates previously ignored properties of qualitative coding such as the sequence in which passages are annotated, the importance of rare codes and the differences in annotation styles between coders. We investigate the QCS task by releasing the first publicly available qualitative coding dataset, CVDQuoding, consisting of interviews conducted with women at risk of cardiovascular disease. In addition, we conduct a human evaluation which shows that our systems consistently make relevant code suggestions.

Our submission to the CRAC 2023 shared task, described herein, is an adapted entity-ranking model jointly trained on all 17 datasets spanning 12 languages. Our model outperforms the shared task baselines by a difference in F1 score of +8.47, achieving an ultimate F1 score of 65.43 and fourth place in the shared task. We explore design decisions related to data preprocessing, the pretrained encoder, and data mixing.

Abstractive summarization systems aim to write concise summaries capturing the most essential information of the input document in their own words. One of the ways to achieve this is to gather and combine multiple pieces of information from the source document, a process we call aggregation. Despite its importance, the extent to which both reference summaries in benchmark datasets and system-generated summaries require aggregation is yet unknown. In this work, we propose AggSHAP, a measure of the degree of aggregation in a summary sentence. We show that AggSHAP distinguishes multi-sentence aggregation from single-sentence extraction or paraphrasing through automatic and human evaluations. We find that few reference or model-generated summary sentences have a high degree of aggregation measured by the proposed metric. We also demonstrate negative correlations between AggSHAP and other quality scores of system summaries. These findings suggest the need to develop new tasks and datasets to encourage multi-sentence aggregation in summarization.

Transformer models pre-trained with a masked-language-modeling objective (e.g., BERT) encode commonsense knowledge as evidenced by behavioral probes; however, the extent to which this knowledge is acquired by systematic inference over the semantics of the pre-training corpora is an open question. To answer this question, we selectively inject verbalized knowledge into the pre-training minibatches of BERT and evaluate how well the model generalizes to supported inferences after pre-training on the injected knowledge. We find generalization does not improve over the course of pre-training BERT from scratch, suggesting that commonsense knowledge is acquired from surface-level, co-occurrence patterns rather than induced, systematic reasoning.

State-of-the-art abstractive summarization systems often generate hallucinations; i.e., content that is not directly inferable from the source text. Despite being assumed to be incorrect, we find that much hallucinated content is actually consistent with world knowledge, which we call factual hallucinations. Including these factual hallucinations in a summary can be beneficial because they provide useful background information. In this work, we propose a novel detection approach that separates factual from non-factual hallucinations of entities. Our method is based on an entity’s prior and posterior probabilities according to pre-trained and finetuned masked language models, respectively. Empirical results suggest that our method vastly outperforms two baselines in both accuracy and F1 scores and has a strong correlation with human judgments on factuality classification tasks. Furthermore, we use our method as a reward signal to train a summarization system using an off-line reinforcement learning (RL) algorithm that can significantly improve the factuality of generated summaries while maintaining the level of abstractiveness.

Idioms are unlike most phrases in two important ways. First, words in an idiom have non-canonical meanings. Second, the non-canonical meanings of words in an idiom are contingent on the presence of other words in the idiom. Linguistic theories differ on whether these properties depend on one another, as well as whether special theoretical machinery is needed to accommodate idioms. We define two measures that correspond to the properties above, and we show that idioms fall at the expected intersection of the two dimensions, but that the dimensions themselves are not correlated. Our results suggest that introducing special machinery to handle idioms may not be warranted.

Authorship obfuscation techniques have commonly been evaluated based on their ability to hide the author’s identity (evasion) while preserving the content of the original text. However, to avoid overstating the systems’ effectiveness, evasion detection must be evaluated using competitive identification techniques in settings that mimic real-life scenarios, and the outcomes of the content-preservation evaluation have to be interpretable by potential users of these obfuscation tools. Motivated by recent work on cross-topic authorship identification and content preservation in summarization, we re-evaluate different authorship obfuscation techniques on detection evasion and content preservation. Furthermore, we propose a new information-theoretic measure to characterize the misattribution harm that can be caused by detection evasion. Our results reveal key weaknesses in state-of-the-art obfuscation techniques and a surprisingly competitive effectiveness from a back-translation baseline in all evaluation aspects.

State-of-the-art abstractive summarization systems frequently hallucinate content that is not supported by the source document, mainly due to noise in the training dataset.Existing methods opt to drop the noisy samples or tokens from the training set entirely, reducing the effective training set size and creating an artificial propensity to copy words from the source. In this work, we propose a training objective for abstractive summarization based on rejection learning, in which the model learns whether or not to reject potentially noisy tokens. We further propose a regularized decoding objective that penalizes non-factual candidate summaries during inference by using the rejection probability learned during training.We show that our method considerably improves the factuality of generated summaries in automatic and human evaluations when compared to five baseline models, and that it does so while increasing the abstractiveness of the generated summaries.

Current language generation models suffer from issues such as repetition, incoherence, and hallucinations. An often-repeated hypothesis for this brittleness of generation models is that it is caused by the training and the generation procedure mismatch, also referred to as exposure bias. In this paper, we verify this hypothesis by analyzing exposure bias from an imitation learning perspective. We show that exposure bias leads to an accumulation of errors during generation, analyze why perplexity fails to capture this accumulation of errors, and empirically show that this accumulation results in poor generation quality.

Most research on question answering focuses on the pre-deployment stage; i.e., building an accurate model for deployment. In this paper, we ask the question: Can we improve QA systems further post-deployment based on user interactions? We focus on two kinds of improvements: 1) improving the QA system’s performance itself, and 2) providing the model with the ability to explain the correctness or incorrectness of an answer. We collect a retrieval-based QA dataset, FeedbackQA, which contains interactive feedback from users. We collect this dataset by deploying a base QA system to crowdworkers who then engage with the system and provide feedback on the quality of its answers. The feedback contains both structured ratings and unstructured natural language explanations. We train a neural model with this feedback data that can generate explanations and re-score answer candidates. We show that feedback data not only improves the accuracy of the deployed QA system but also other stronger non-deployed systems. The generated explanations also help users make informed decisions about the correctness of answers.

Presuppositions are assumptions that are taken for granted by an utterance, and identifying them is key to a pragmatic interpretation of language. In this paper, we investigate the capabilities of transformer models to perform NLI on cases involving presupposition. First, we present simple heuristics to create alternative “contrastive” test cases based on the ImpPres dataset and investigate the model performance on those test cases. Second, to better understand how the model is making its predictions, we analyze samples from sub-datasets of ImpPres and examine model performance on them. Overall, our findings suggest that NLI-trained transformer models seem to be exploiting specific structural and lexical cues as opposed to performing some kind of pragmatic reasoning.

In a text, entities mentioned earlier can be referred to in later discourse by a more general description. For example, Celine Dion and Justin Bieber can be referred to by Canadian singers or celebrities. In this work, we study this phenomenon in the context of summarization, where entities from a source text are generalized in the summary. We call such instances source-summary entity aggregations. We categorize these aggregations into two types and analyze them in the Cnn/Dailymail corpus, showing that they are reasonably frequent. We then examine how well three state-of-the-art summarization systems can generate such aggregations within summaries. We also develop techniques to encourage them to generate more aggregations. Our results show that there is significant room for improvement in producing semantically correct aggregations.

It is a common belief that training deep transformers from scratch requires large datasets. Consequently, for small datasets, people usually use shallow and simple additional layers on top of pre-trained models during fine-tuning. This work shows that this does not always need to be the case: with proper initialization and optimization, the benefits of very deep transformers can carry over to challenging tasks with small datasets, including Text-to-SQL semantic parsing and logical reading comprehension. In particular, we successfully train 48 layers of transformers, comprising 24 fine-tuned layers from pre-trained RoBERTa and 24 relation-aware layers trained from scratch. With fewer training steps and no task-specific pre-training, we obtain the state of the art performance on the challenging cross-domain Text-to-SQL parsing benchmark Spider. We achieve this by deriving a novel Data dependent Transformer Fixed-update initialization scheme (DT-Fixup), inspired by the prior T-Fixup work. Further error analysis shows that increasing depth can help improve generalization on small datasets for hard cases that require reasoning and structural understanding.

A false contract is more likely to be rejected than a contract is, yet a false key is less likely than a key to open doors. While correctly interpreting and assessing the effects of such adjective-noun pairs (e.g., false key) on the plausibility of given events (e.g., opening doors) underpins many natural language understanding tasks, doing so often requires a significant degree of world knowledge and common-sense reasoning. We introduce ADEPT – a large-scale semantic plausibility task consisting of over 16 thousand sentences that are paired with slightly modified versions obtained by adding an adjective to a noun. Overall, we find that while the task appears easier for human judges (85% accuracy), it proves more difficult for transformer-based models like RoBERTa (71% accuracy). Our experiments also show that neither the adjective itself nor its taxonomic class suffice in determining the correct plausibility judgement, emphasizing the importance of endowing automatic natural language understanding systems with more context sensitivity and common-sense reasoning.

Understanding natural language requires common sense, one aspect of which is the ability to discern the plausibility of events. While distributional models—most recently pre-trained, Transformer language models—have demonstrated improvements in modeling event plausibility, their performance still falls short of humans’. In this work, we show that Transformer-based plausibility models are markedly inconsistent across the conceptual classes of a lexical hierarchy, inferring that “a person breathing” is plausible while “a dentist breathing” is not, for example. We find this inconsistency persists even when models are softly injected with lexical knowledge, and we present a simple post-hoc method of forcing model consistency that improves correlation with human plausibility judgements.

We propose an unsupervised graph-based ranking model for extractive summarization of long scientific documents. Our method assumes a two-level hierarchical graph representation of the source document, and exploits asymmetrical positional cues to determine sentence importance. Results on the PubMed and arXiv datasets show that our approach outperforms strong unsupervised baselines by wide margins in automatic metrics and human evaluation. In addition, it achieves performance comparable to many state-of-the-art supervised approaches which are trained on hundreds of thousands of examples. These results suggest that patterns in the discourse structure are a strong signal for determining importance in scientific articles.

Extractive summarization has been the mainstay of automatic summarization for decades. Despite all the progress, extractive summarizers still suffer from shortcomings including coreference issues arising from extracting sentences away from their original context in the source document. This affects the coherence and readability of extractive summaries. In this work, we propose a lightweight post-editing step for extractive summaries that centers around a single linguistic decision: the definiteness of noun phrases. We conduct human evaluation studies that show that human expert judges substantially prefer the output of our proposed system over the original summaries. Moreover, based on an automatic evaluation study, we provide evidence for our system’s ability to generate linguistic decisions that lead to improved extractive summaries. We also draw insights about how the automatic system is exploiting some local cues related to the writing style of the main article texts or summary texts to make the decisions, rather than reasoning about the contexts pragmatically.

Natural language processing systems such as dialogue agents should be able to reason about other people’s beliefs, intentions and desires. This capability, called theory of mind (ToM), is crucial, as it allows a model to predict and interpret the needs of users based on their mental states. A recent line of research evaluates the ToM capability of existing memory-augmented neural models through question-answering. These models perform poorly on false belief tasks where beliefs differ from reality, especially when the dataset contains distracting sentences. In this paper, we propose a new temporally informed approach for improving the ToM capability of memory-augmented neural models. Our model incorporates priors about the entities’ minds and tracks their mental states as they evolve over time through an extended passage. It then responds to queries through textual time travel–i.e., by accessing the stored memory of an earlier time step. We evaluate our model on ToM datasets and find that this approach improves performance, particularly by correcting the predicted mental states to match the false belief.

Authorship attribution is the problem of identifying the most plausible author of an anonymous text from a set of candidate authors. Researchers have investigated same-topic and cross-topic scenarios of authorship attribution, which differ according to whether new, unseen topics are used in the testing phase. However, neither scenario allows us to explain whether errors are caused by failure to capture authorship writing style or by the topic shift. Motivated by this, we propose the topic confusion task where we switch the author-topic configuration between the training and testing sets. This setup allows us to investigate two types of errors: one caused by the topic shift and one caused by the features’ inability to capture the writing styles. We show that stylometric features with part-of-speech tags are the least susceptible to topic variations. We further show that combining them with other features leads to significantly lower topic confusion and higher attribution accuracy. Finally, we show that pretrained language models such as BERT and RoBERTa perform poorly on this task and are surpassed by simple features such as word-level n-gram.

Contextualized word representations have become a driving force in NLP, motivating widespread interest in understanding their capabilities and the mechanisms by which they operate. Particularly intriguing is their ability to identify and encode conceptual abstractions. Past work has probed BERT representations for this competence, finding that BERT can correctly retrieve noun hypernyms in cloze tasks. In this work, we ask the question: do probing studies shed light on systematic knowledge in BERT representations? As a case study, we examine hypernymy knowledge encoded in BERT representations. In particular, we demonstrate through a simple consistency probe that the ability to correctly retrieve hypernyms in cloze tasks, as used in prior work, does not correspond to systematic knowledge in BERT. Our main conclusion is cautionary: even if BERT demonstrates high probing accuracy for a particular competence, it does not necessarily follow that BERT ‘understands’ a concept, and it cannot be expected to systematically generalize across applicable contexts.

Word embeddings are trained to predict word cooccurrence statistics, which leads them to possess different lexical properties (syntactic, semantic, etc.) depending on the notion of context defined at training time. These properties manifest when querying the embedding space for the most similar vectors, and when used at the input layer of deep neural networks trained to solve downstream NLP problems. Meta-embeddings combine multiple sets of differently trained word embeddings, and have been shown to successfully improve intrinsic and extrinsic performance over equivalent models which use just one set of source embeddings. We introduce word prisms: a simple and efficient meta-embedding method that learns to combine source embeddings according to the task at hand. Word prisms learn orthogonal transformations to linearly combine the input source embeddings, which allows them to be very efficient at inference time. We evaluate word prisms in comparison to other meta-embedding methods on six extrinsic evaluations and observe that word prisms offer improvements in performance on all tasks.

The Winograd Schema Challenge (WSC) and variants inspired by it have become important benchmarks for common-sense reasoning (CSR). Model performance on the WSC has quickly progressed from chance-level to near-human using neural language models trained on massive corpora. In this paper, we analyze the effects of varying degrees of overlaps that occur between these corpora and the test instances in WSC-style tasks. We find that a large number of test instances overlap considerably with the pretraining corpora on which state-of-the-art models are trained, and that a significant drop in classification accuracy occurs when models are evaluated on instances with minimal overlap. Based on these results, we provide the WSC-Web dataset, consisting of over 60k pronoun disambiguation problems scraped from web data, being both the largest corpus to date, and having a significantly lower proportion of overlaps with current pretraining corpora.

In this paper, we propose LexSub, a novel approach towards unifying lexical and distributional semantics. We inject knowledge about lexical-semantic relations into distributional word embeddings by defining subspaces of the distributional vector space in which a lexical relation should hold. Our framework can handle symmetric attract and repel relations (e.g., synonymy and antonymy, respectively), as well as asymmetric relations (e.g., hypernymy and meronomy). In a suite of intrinsic benchmarks, we show that our model outperforms previous approaches on relatedness tasks and on hypernymy classification and detection, while being competitive on word similarity tasks. It also outperforms previous systems on extrinsic classification tasks that benefit from exploiting lexical relational cues. We perform a series of analyses to understand the behaviors of our model.1Code available at https://github.com/aishikchakraborty/LexSub.

Inferring missing facts in temporal knowledge graphs (TKGs) is a fundamental and challenging task. Previous works have approached this problem by augmenting methods for static knowledge graphs to leverage time-dependent representations. However, these methods do not explicitly leverage multi-hop structural information and temporal facts from recent time steps to enhance their predictions. Additionally, prior work does not explicitly address the temporal sparsity and variability of entity distributions in TKGs. We propose the Temporal Message Passing (TeMP) framework to address these challenges by combining graph neural networks, temporal dynamics models, data imputation and frequency-based gating techniques. Experiments on standard TKG tasks show that our approach provides substantial gains compared to the previous state of the art, achieving a 10.7% average relative improvement in Hits@10 across three standard benchmarks. Our analysis also reveals important sources of variability both within and across TKG datasets, and we introduce several simple but strong baselines that outperform the prior state of the art in certain settings.

Neural abstractive summarization systems have achieved promising progress, thanks to the availability of large-scale datasets and models pre-trained with self-supervised methods. However, ensuring the factual consistency of the generated summaries for abstractive summarization systems is a challenge. We propose a post-editing corrector module to address this issue by identifying and correcting factual errors in generated summaries. The neural corrector model is pre-trained on artificial examples that are created by applying a series of heuristic transformations on reference summaries. These transformations are inspired by the error analysis of state-of-the-art summarization model outputs. Experimental results show that our model is able to correct factual errors in summaries generated by other neural summarization models and outperforms previous models on factual consistency evaluation on the CNN/DailyMail dataset. We also find that transferring from artificial error correction to downstream settings is still very challenging.

Human-written texts contain frequent generalizations and semantic aggregation of content. In a document, they may refer to a pair of named entities such as ‘London’ and ‘Paris’ with different expressions: “the major cities”, “the capital cities” and “two European cities”. Yet generation, especially, abstractive summarization systems have so far focused heavily on paraphrasing and simplifying the source content, to the exclusion of such semantic abstraction capabilities. In this paper, we present a new dataset and task aimed at the semantic aggregation of entities. TESA contains a dataset of 5.3K crowd-sourced entity aggregations of Person, Organization, and Location named entities. The aggregations are document-appropriate, meaning that they are produced by annotators to match the situational context of a given news article from the New York Times. We then build baseline models for generating aggregations given a tuple of entities and document context. We finetune on TESA an encoder-decoder language model and compare it with simpler classification methods based on linguistically informed features. Our quantitative and qualitative evaluations show reasonable performance in making a choice from a given list of expressions, but free-form expressions are understandably harder to generate and evaluate.

Word embeddings are reliable feature representations of words used to obtain high quality results for various NLP applications. Uncontextualized word embeddings are used in many NLP tasks today, especially in resource-limited settings where high memory capacity and GPUs are not available. Given the historical success of word embeddings in NLP, we propose a retrospective on some of the most well-known word embedding algorithms. In this work, we deconstruct Word2vec, GloVe, and others, into a common form, unveiling some of the common conditions that seem to be required for making performant word embeddings. We believe that the theoretical findings in this paper can provide a basis for more informed development of future models.

Pre-trained neural abstractive summarization systems have dominated extractive strategies on news summarization performance, at least in terms of ROUGE. However, system-generated abstractive summaries often face the pitfall of factual inconsistency: generating incorrect facts with respect to the source text. To address this challenge, we propose Span-Fact, a suite of two factual correction models that leverages knowledge learned from question answering models to make corrections in system-generated summaries via span selection. Our models employ single or multi-masking strategies to either iteratively or auto-regressively replace entities in order to ensure semantic consistency w.r.t. the source text, while retaining the syntactic structure of summaries generated by abstractive summarization models. Experiments show that our models significantly boost the factual consistency of system-generated summaries without sacrificing summary quality in terms of both automatic metrics and human evaluation.

Coherence is an important aspect of text quality and is crucial for ensuring its readability. One important limitation of existing coherence models is that training on one domain does not easily generalize to unseen categories of text. Previous work advocates for generative models for cross-domain generalization, because for discriminative models, the space of incoherent sentence orderings to discriminate against during training is prohibitively large. In this work, we propose a local discriminative neural model with a much smaller negative sampling space that can efficiently learn against incorrect orderings. The proposed coherence model is simple in structure, yet it significantly outperforms previous state-of-art methods on a standard benchmark dataset on the Wall Street Journal corpus, as well as in multiple new challenging settings of transfer to unseen categories of discourse on Wikipedia articles.

We present the first sentence simplification model that learns explicit edit operations (ADD, DELETE, and KEEP) via a neural programmer-interpreter approach. Most current neural sentence simplification systems are variants of sequence-to-sequence models adopted from machine translation. These methods learn to simplify sentences as a byproduct of the fact that they are trained on complex-simple sentence pairs. By contrast, our neural programmer-interpreter is directly trained to predict explicit edit operations on targeted parts of the input sentence, resembling the way that humans perform simplification and revision. Our model outperforms previous state-of-the-art neural sentence simplification models (without external knowledge) by large margins on three benchmark text simplification corpora in terms of SARI (+0.95 WikiLarge, +1.89 WikiSmall, +1.41 Newsela), and is judged by humans to produce overall better and simpler output sentences.

We introduce a new benchmark for coreference resolution and NLI, KnowRef, that targets common-sense understanding and world knowledge. Previous coreference resolution tasks can largely be solved by exploiting the number and gender of the antecedents, or have been handcrafted and do not reflect the diversity of naturally occurring text. We present a corpus of over 8,000 annotated text passages with ambiguous pronominal anaphora. These instances are both challenging and realistic. We show that various coreference systems, whether rule-based, feature-rich, or neural, perform significantly worse on the task than humans, who display high inter-annotator agreement. To explain this performance gap, we show empirically that state-of-the art models often fail to capture context, instead relying on the gender or number of candidate antecedents to make a decision. We then use problem-specific insights to propose a data-augmentation trick called antecedent switching to alleviate this tendency in models. Finally, we show that antecedent switching yields promising results on other tasks as well: we use it to achieve state-of-the-art results on the GAP coreference task.

Neural abstractive summarizers generate summary texts using a language model conditioned on the input source text, and have recently achieved high ROUGE scores on benchmark summarization datasets. We investigate how they achieve this performance with respect to human-written gold-standard abstracts, and whether the systems are able to understand deeper syntactic and semantic structures. We generate a set of contrastive summaries which are perturbed, deficient versions of human-written summaries, and test whether existing neural summarizers score them more highly than the human-written summaries. We analyze their performance on different datasets and find that these systems fail to understand the source text, in a majority of the cases.

Referring Expression Generation (REG) is the task of generating contextually appropriate references to entities. A limitation of existing REG systems is that they rely on entity-specific supervised training, which means that they cannot handle entities not seen during training. In this study, we address this in two ways. First, we propose task setups in which we specifically test a REG system’s ability to generalize to entities not seen during training. Second, we propose a profile-based deep neural network model, ProfileREG, which encodes both the local context and an external profile of the entity to generate reference realizations. Our model generates tokens by learning to choose between generating pronouns, generating from a fixed vocabulary, or copying a word from the profile. We evaluate our model on three different splits of the WebNLG dataset, and show that it outperforms competitive baselines in all settings according to automatic and human evaluations.

Recent studies have significantly improved the state-of-the-art on common-sense reasoning (CSR) benchmarks like the Winograd Schema Challenge (WSC) and SWAG. The question we ask in this paper is whether improved performance on these benchmarks represents genuine progress towards common-sense-enabled systems. We make case studies of both benchmarks and design protocols that clarify and qualify the results of previous work by analyzing threats to the validity of previous experimental designs. Our protocols account for several properties prevalent in common-sense benchmarks including size limitations, structural regularities, and variable instance difficulty.

Sentence position is a strong feature for news summarization, since the lead often (but not always) summarizes the key points of the article. In this paper, we show that recent neural systems excessively exploit this trend, which although powerful for many inputs, is also detrimental when summarizing documents where important content should be extracted from later parts of the article. We propose two techniques to make systems sensitive to the importance of content in different parts of the article. The first technique employs ‘unbiased’ data; i.e., randomly shuffled sentences of the source document, to pretrain the model. The second technique uses an auxiliary ROUGE-based loss that encourages the model to distribute importance scores throughout a document by mimicking sentence-level ROUGE scores on the training data. We show that these techniques significantly improve the performance of a competitive reinforcement learning based extractive system, with the auxiliary loss being more powerful than pretraining.

Modeling semantic plausibility requires commonsense knowledge about the world and has been used as a testbed for exploring various knowledge representations. Previous work has focused specifically on modeling physical plausibility and shown that distributional methods fail when tested in a supervised setting. At the same time, distributional models, namely large pretrained language models, have led to improved results for many natural language understanding tasks. In this work, we show that these pretrained language models are in fact effective at modeling physical plausibility in the supervised setting. We therefore present the more difficult problem of learning to model physical plausibility directly from text. We create a training set by extracting attested events from a large corpus, and we provide a baseline for training on these attested events in a self-supervised manner and testing on a physical plausibility task. We believe results could be further improved by injecting explicit commonsense knowledge into a distributional model.

We introduce an automatic system that performs well on two common-sense reasoning tasks, the Winograd Schema Challenge (WSC) and the Choice of Plausible Alternatives (COPA). Problem instances from these tasks require diverse, complex forms of inference and knowledge to solve. Our method uses a knowledge-hunting module to gather text from the web, which serves as evidence for candidate problem resolutions. Given an input problem, our system generates relevant queries to send to a search engine. It extracts and classifies knowledge from the returned results and weighs it to make a resolution. Our approach improves F1 performance on the WSC by 0.16 over the previous best and is competitive with the state-of-the-art on COPA, demonstrating its general applicability.

We introduce the novel task of predicting adverbial presupposition triggers, which is useful for natural language generation tasks such as summarization and dialogue systems. We introduce two new corpora, derived from the Penn Treebank and the Annotated English Gigaword dataset and investigate the use of a novel attention mechanism tailored to this task. Our attention mechanism augments a baseline recurrent neural network without the need for additional trainable parameters, minimizing the added computational cost of our mechanism. We demonstrate that this model statistically outperforms our baselines.

Commonsense knowledge bases such as ConceptNet represent knowledge in the form of relational triples. Inspired by recent work by Li et al., we analyse if knowledge base completion models can be used to mine commonsense knowledge from raw text. We propose novelty of predicted triples with respect to the training set as an important factor in interpreting results. We critically analyse the difficulty of mining novel commonsense knowledge, and show that a simple baseline method that outperforms the previous state of the art on predicting more novel triples.

Deep neural networks have been displaying superior performance over traditional supervised classifiers in text classification. They learn to extract useful features automatically when sufficient amount of data is presented. However, along with the growth in the number of documents comes the increase in the number of categories, which often results in poor performance of the multiclass classifiers. In this work, we use external knowledge in the form of topic category taxonomies to aide the classification by introducing a deep hierarchical neural attention-based classifier. Our model performs better than or comparable to state-of-the-art hierarchical models at significantly lower computational cost while maintaining high interpretability.

We introduce an automatic system that achieves state-of-the-art results on the Winograd Schema Challenge (WSC), a common sense reasoning task that requires diverse, complex forms of inference and knowledge. Our method uses a knowledge hunting module to gather text from the web, which serves as evidence for candidate problem resolutions. Given an input problem, our system generates relevant queries to send to a search engine, then extracts and classifies knowledge from the returned results and weighs them to make a resolution. Our approach improves F1 performance on the full WSC by 0.21 over the previous best and represents the first system to exceed 0.5 F1. We further demonstrate that the approach is competitive on the Choice of Plausible Alternatives (COPA) task, which suggests that it is generally applicable.

In this work, we propose a novel method for training neural networks to perform single-document extractive summarization without heuristically-generated extractive labels. We call our approach BanditSum as it treats extractive summarization as a contextual bandit (CB) problem, where the model receives a document to summarize (the context), and chooses a sequence of sentences to include in the summary (the action). A policy gradient reinforcement learning algorithm is used to train the model to select sequences of sentences that maximize ROUGE score. We perform a series of experiments demonstrating that BanditSum is able to achieve ROUGE scores that are better than or comparable to the state-of-the-art for extractive summarization, and converges using significantly fewer update steps than competing approaches. In addition, we show empirically that BanditSum performs significantly better than competing approaches when good summary sentences appear late in the source document.

We present an approach to event coreference resolution by developing a general framework for clustering that uses supervised representation learning. We propose a neural network architecture with novel Clustering-Oriented Regularization (CORE) terms in the objective function. These terms encourage the model to create embeddings of event mentions that are amenable to clustering. We then use agglomerative clustering on these embeddings to build event coreference chains. For both within- and cross-document coreference on the ECB+ corpus, our model obtains better results than models that require significantly more pre-annotated information. This work provides insight and motivating results for a new general approach to solving coreference and clustering problems with representation learning.

We present a method for unsupervised lexical frame acquisition at the syntax–semantics interface. Given a set of input strings derived from dependency parses, our method generates a set of clusters that resemble lexical frame structures. Our work is motivated not only by its practical applications (e.g., to build, or expand the coverage of lexical frame databases), but also to gain linguistic insight into frame structures with respect to lexical distributions in relation to grammatical structures. We model our task using a hierarchical Bayesian network and employ tools and methods from latent variable probabilistic context free grammars (L-PCFGs) for statistical inference and parameter fitting, for which we propose a new split and merge procedure. We show that our model outperforms several baselines on a portion of the Wall Street Journal sentences that we have newly annotated for evaluation purposes.

Humans interpret texts with respect to some background information, or world knowledge, and we would like to develop automatic reading comprehension systems that can do the same. In this paper, we introduce a task and several models to drive progress towards this goal. In particular, we propose the task of rare entity prediction: given a web document with several entities removed, models are tasked with predicting the correct missing entities conditioned on the document context and the lexical resources. This task is challenging due to the diversity of language styles and the extremely large number of rare entities. We propose two recurrent neural network architectures which make use of external knowledge in the form of entity descriptions. Our experiments show that our hierarchical LSTM model performs significantly better at the rare entity prediction task than those that do not make use of external resources.

In goal-driven dialogue systems, success is often defined based on a structured definition of the goal. This requires that the dialogue system be constrained to handle a specific class of goals and that there be a mechanism to measure success with respect to that goal. However, in many human-human dialogues the diversity of goals makes it infeasible to define success in such a way. To address this scenario, we consider the task of automatically predicting success in goal-driven human-human dialogues using only the information communicated between participants in the form of text. We build a dataset from stackoverflow.com which consists of exchanges between two users in the technical domain where ground-truth success labels are available. We then propose a turn-based hierarchical neural network model that can be used to predict success without requiring a structured goal definition. We show this model outperforms rule-based heuristics and other baselines as it is able to detect patterns over the course of a dialogue and capture notions such as gratitude.

We examine the task of aggregation in the context of text-to-text generation. We introduce a new aggregation task which frames the process as grouping input sentence fragments into clusters that are to be expressed as a single output sentence. We extract datasets for this task from a corpus using an automatic extraction process. Based on the results of a user study, we develop two gold-standard clusterings and corresponding evaluation methods for each dataset. We present a hierarchical clustering framework for predicting aggregation decisions on this task, which outperforms several baselines and can serve as a reference in future work.

We examine the potential of recurrent neural networks for handling pragmatic inferences involving complex contextual cues for the task of article usage prediction. We train and compare several variants of Long Short-Term Memory (LSTM) networks with an attention mechanism. Our model outperforms a previous state-of-the-art system, achieving up to 96.63% accuracy on the WSJ/PTB corpus. In addition, we perform a series of analyses to understand the impact of various model choices. We find that the gain in performance can be attributed to the ability of LSTMs to pick up on contextual cues, both local and further away in distance, and that the model is able to solve cases involving reasoning about coreference and synonymy. We also show how the attention mechanism contributes to the interpretability of the model’s effectiveness.