While large language models (LLMs) have proven to be effective on a large variety of tasks, they are also known to hallucinate information. To measure whether an LLM prefers factually consistent continuations of its input, we propose a new benchmark called FIB (Factual Inconsistency Benchmark) that focuses on the task of summarization. Specifically, our benchmark involves comparing the scores an LLM assigns to a factually consistent versus a factually inconsistent summary for an input news article. For factually consistent summaries, we use human-written reference summaries that we manually verify as factually consistent. To generate summaries that are factually inconsistent, we generate summaries from a suite of summarization models that we have manually annotated as factually inconsistent. A model’s factual consistency is then measured according to its accuracy, i.e. the proportion of documents where it assigns a higher score to the factually consistent summary. To validate the usefulness of {pasted macro ‘BENCHMARK’}, we evaluate 23 large language models ranging from 1B to 176B parameters from six different model families including BLOOM and OPT. We find that existing LLMs generally assign a higher score to factually consistent summaries than to factually inconsistent summaries. However, if the factually inconsistent summaries occur verbatim in the document, then LLMs assign a higher score to these factually inconsistent summaries than factually consistent summaries. We validate design choices in our benchmark including the scoring method and source of distractor summaries.

Language models (LMs) can generate hallucinations and incoherent outputs, which highlights their weak context dependency. Cache-LMs, which augment LMs with a memory of recent history, can increase context dependency and have shown remarkable performance in diverse language generation tasks. However, we find that even with training, the performance gain stemming from the cache component of current cache-LMs is suboptimal due to the misalignment between the current hidden states and those stored in the memory. In this work, we present HistAlign, a new training approach to ensure good cache alignment such that the model receives useful signals from the history. We first prove our concept on a simple and synthetic task where the memory is essential for correct predictions, and we show that the cache component of HistAlign is better aligned and improves overall performance. Next, we evaluate HistAlign on diverse downstream language generation tasks, including prompt continuation, abstractive summarization, and data-to-text. We demonstrate that HistAlign improves text coherence and faithfulness in open-ended and conditional generation settings respectively. HistAlign is also generalizable across different model families, showcasing its strength in improving context dependency of LMs in diverse scenarios.

The problems of unfaithful summaries have been widely discussed under the context of abstractive summarization. Though extractive summarization is less prone to the common unfaithfulness issues of abstractive summaries, does that mean extractive is equal to faithful? Turns out that the answer is no. In this work, we define a typology with five types of broad unfaithfulness problems (including and beyond not-entailment) that can appear in extractive summaries, including incorrect coreference, incomplete coreference, incorrect discourse, incomplete discourse, as well as other misleading information. We ask humans to label these problems out of 1600 English summaries produced by 16 diverse extractive systems. We find that 30% of the summaries have at least one of the five issues. To automatically detect these problems, we find that 5 existing faithfulness evaluation metrics for summarization have poor correlations with human judgment. To remedy this, we propose a new metric, ExtEval, that is designed for detecting unfaithful extractive summaries and is shown to have the best performance. We hope our work can increase the awareness of unfaithfulness problems in extractive summarization and help future work to evaluate and resolve these issues.

Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P – that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may “over-generalize”, in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective that mixes the forward and reverse cross-entropies. We evaluate models trained with this objective on synthetic data settings (where P is known) and real data, and show that the resulting models yield better generated text without complex decoding strategies.

A multilingual tokenizer is a fundamental component of multilingual neural machine translation. It is trained from a multilingual corpus. Since a skewed data distribution is considered to be harmful, a sampling strategy is usually used to balance languages in the corpus. However, few works have systematically answered how language imbalance in tokenizer training affects downstream performance. In this work, we analyze how translation performance changes as the data ratios among languages vary in the tokenizer training corpus. We find that while relatively better performance is often observed when languages are more equally sampled, the downstream performance is more robust to language imbalance than we usually expected. Two features, UNK rate and closeness to the character level, can warn of poor downstream performance before performing the task. We also distinguish language sampling for tokenizer training from sampling for model training and show that the model is more sensitive to the latter.

More than 43% of the languages spoken in the world are endangered, and language loss currently occurs at an accelerated rate because of globalization and neocolonialism. Saving and revitalizing endangered languages has become very important for maintaining the cultural diversity on our planet. In this work, we focus on discussing how NLP can help revitalize endangered languages. We first suggest three principles that may help NLP practitioners to foster mutual understanding and collaboration with language communities, and we discuss three ways in which NLP can potentially assist in language education. We then take Cherokee, a severely-endangered Native American language, as a case study. After reviewing the language’s history, linguistic features, and existing resources, we (in collaboration with Cherokee community members) arrive at a few meaningful ways NLP practitioners can collaborate with community partners. We suggest two approaches to enrich the Cherokee language’s resources with machine-in-the-loop processing, and discuss several NLP tools that people from the Cherokee community have shown interest in. We hope that our work serves not only to inform the NLP community about Cherokee, but also to provide inspiration for future work on endangered languages in general.

Previous Part-Of-Speech (POS) induction models usually assume certain independence assumptions (e.g., Markov, unidirectional, local dependency) that do not hold in real languages. For example, the subject-verb agreement can be both long-term and bidirectional. To facilitate flexible dependency modeling, we propose a Masked Part-of-Speech Model (MPoSM), inspired by the recent success of Masked Language Models (MLM). MPoSM can model arbitrary tag dependency and perform POS induction through the objective of masked POS reconstruction. We achieve competitive results on both the English Penn WSJ dataset as well as the universal treebank containing 10 diverse languages. Though modeling the long-term dependency should ideally help this task, our ablation study shows mixed trends in different languages. To better understand this phenomenon, we design a novel synthetic experiment that can specifically diagnose the model’s ability to learn tag agreement. Surprisingly, we find that even strong baselines fail to solve this problem consistently in a very simplified setting: the agreement between adjacent words. Nonetheless, MPoSM achieves overall better performance. Lastly, we conduct a detailed error analysis to shed light on other remaining challenges.

Student Evaluations of Teaching (SETs) are widely used in colleges and universities. Typically SET results are summarized for instructors in a static PDF report. The report often includes summary statistics for quantitative ratings and an unsorted list of open-ended student comments. The lack of organization and summarization of the raw comments hinders those interpreting the reports from fully utilizing informative feedback, making accurate inferences, and designing appropriate instructional improvements. In this work, we introduce a novel system, SETSUM, that leverages sentiment analysis, aspect extraction, summarization, and visualization techniques to provide organized illustrations of SET findings to instructors and other reviewers. Ten university professors from diverse departments serve as evaluators of the system and all agree that SETSUM help them interpret SET results more efficiently; and 6 out of 10 instructors prefer our system over the standard static PDF report (while the remaining 4 would like to have both). This demonstrates that our work holds the potential of reforming the SET reporting conventions in the future.

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.

Recent years have witnessed various types of generative models for natural language generation (NLG), especially RNNs or transformer based sequence-to-sequence models, as well as variational autoencoder (VAE) and generative adversarial network (GAN) based models. However, flow-based generative models, which achieve strong performance in image generation due to their invertibility and exact density estimation properties, have been less explored for NLG. In this paper, we propose a flow-based language generation model by adapting previous flow generative models to language generation via continuous input embeddings, adapted affine coupling structures, and a novel architecture for autoregressive text generation. We also apply our framework to Sequence-to-Sequence generation, including text- and video-based Question Generation (QG) and Neural Machine Translation (NMT), and data augmentation for Question Answering (QA). We use our language flow model to provide extra input features for QG and NMT, which achieves improvements over the strong QG baselines on SQuAD and TVQA and NMT baseline on WMT16. We also augment QA data with new context by injecting noise to the latent features of the language flow and show this augmentation leads to a large performance improvement from strong baselines on SQuAD and TVQA.

Recent years have brought about an interest in the challenging task of summarizing conversation threads (meetings, online discussions, etc.). Such summaries help analysis of the long text to quickly catch up with the decisions made and thus improve our work or communication efficiency. To spur research in thread summarization, we have developed an abstractive Email Thread Summarization (EmailSum) dataset, which contains human-annotated short (<30 words) and long (<100 words) summaries of 2,549 email threads (each containing 3 to 10 emails) over a wide variety of topics. We perform a comprehensive empirical study to explore different summarization techniques (including extractive and abstractive methods, single-document and hierarchical models, as well as transfer and semisupervised learning) and conduct human evaluations on both short and long summary generation tasks. Our results reveal the key challenges of current abstractive summarization models in this task, such as understanding the sender’s intent and identifying the roles of sender and receiver. Furthermore, we find that widely used automatic evaluation metrics (ROUGE, BERTScore) are weakly correlated with human judgments on this email thread summarization task. Hence, we emphasize the importance of human evaluation and the development of better metrics by the community.

We introduce ChrEnTranslate, an online machine translation demonstration system for translation between English and an endangered language Cherokee. It supports both statistical and neural translation models as well as provides quality estimation to inform users of reliability, two user feedback interfaces for experts and common users respectively, example inputs to collect human translations for monolingual data, word alignment visualization, and relevant terms from the Cherokee English dictionary. The quantitative evaluation demonstrates that our backbone translation models achieve state-of-the-art translation performance and our quality estimation well correlates with both BLEU and human judgment. By analyzing 216 pieces of expert feedback, we find that NMT is preferable because it copies less than SMT, and, in general, current models can translate fragments of the source sentence but make major mistakes. When we add these 216 expert-corrected parallel texts into the training set and retrain models, equal or slightly better performance is observed, which demonstrates indicates the potential of human-in-the-loop learning.

Data collection for natural language (NL) understanding tasks has increasingly included human explanations alongside data points, allowing past works to introduce models that both perform a task and generate NL explanations for their outputs. Yet to date, model-generated explanations have been evaluated on the basis of surface-level similarities to human explanations, both through automatic metrics like BLEU and human evaluations. We argue that these evaluations are insufficient, since they fail to indicate whether explanations support actual model behavior (faithfulness), rather than simply match what a human would say (plausibility). In this work, we address the problem of evaluating explanations from the the model simulatability perspective. Our contributions are as follows: (1) We introduce a leakage-adjusted simulatability (LAS) metric for evaluating NL explanations, which measures how well explanations help an observer predict a model’s output, while controlling for how explanations can directly leak the output. We use a model as a proxy for a human observer, and validate this choice with two human subject experiments. (2) Using the CoS-E and e-SNLI datasets, we evaluate two existing generative graphical models and two new approaches; one rationalizing method we introduce achieves roughly human-level LAS scores. (3) Lastly, we frame explanation generation as a multi-agent game and optimize explanations for simulatability while penalizing label leakage, which can improve LAS scores.

Cherokee is a highly endangered Native American language spoken by the Cherokee people. The Cherokee culture is deeply embedded in its language. However, there are approximately only 2,000 fluent first language Cherokee speakers remaining in the world and the number is declining every year. To help save this endangered language, we introduce ChrEn, a Cherokee-English parallel dataset, to facilitate machine translation research between Cherokee and English. Compared to some popular machine translation language pairs, ChrEn is extremely low-resource, only containing 14k sentence pairs in total. We split our parallel data in ways that facilitate both in-domain and out-of-domain evaluation. We also collect 5k Cherokee monolingual data to enable semi-supervised learning. Besides these datasets, we propose several Cherokee-English and English-Cherokee machine translation systems. We compare SMT (phrase-based) versus NMT (RNN-based and Transformer-based) systems; supervised versus semi-supervised (via language model, back-translation, and BERT/Multilingual-BERT) methods; as well as transfer learning versus multilingual joint training with 4 other languages. Our best results are 15.8/12.7 BLEU for in-domain and 6.5/5.0 BLEU for out-of-domain Chr-En/EnChr translations, respectively; and we hope that our dataset and systems will encourage future work by the community for Cherokee language revitalization.

Text-based Question Generation (QG) aims at generating natural and relevant questions that can be answered by a given answer in some context. Existing QG models suffer from a “semantic drift” problem, i.e., the semantics of the model-generated question drifts away from the given context and answer. In this paper, we first propose two semantics-enhanced rewards obtained from downstream question paraphrasing and question answering tasks to regularize the QG model to generate semantically valid questions. Second, since the traditional evaluation metrics (e.g., BLEU) often fall short in evaluating the quality of generated questions, we propose a QA-based evaluation method which measures the QG model’s ability to mimic human annotators in generating QA training data. Experiments show that our method achieves the new state-of-the-art performance w.r.t. traditional metrics, and also performs best on our QA-based evaluation metrics. Further, we investigate how to use our QG model to augment QA datasets and enable semi-supervised QA. We propose two ways to generate synthetic QA pairs: generate new questions from existing articles or collect QA pairs from new articles. We also propose two empirically effective strategies, a data filter and mixing mini-batch training, to properly use the QG-generated data for QA. Experiments show that our method improves over both BiDAF and BERT QA baselines, even without introducing new articles.

Neural machine translation (NMT) has achieved notable success in recent times, however it is also widely recognized that this approach has limitations with handling infrequent words and word pairs. This paper presents a novel memory-augmented NMT (M-NMT) architecture, which stores knowledge about how words (usually infrequently encountered ones) should be translated in a memory and then utilizes them to assist the neural model. We use this memory mechanism to combine the knowledge learned from a conventional statistical machine translation system and the rules learned by an NMT system, and also propose a solution for out-of-vocabulary (OOV) words based on this framework. Our experiments on two Chinese-English translation tasks demonstrated that the M-NMT architecture outperformed the NMT baseline by 9.0 and 2.7 BLEU points on the two tasks, respectively. Additionally, we found this architecture resulted in a much more effective OOV treatment compared to competitive methods.

It has been shown that Chinese poems can be successfully generated by sequence-to-sequence neural models, particularly with the attention mechanism. A potential problem of this approach, however, is that neural models can only learn abstract rules, while poem generation is a highly creative process that involves not only rules but also innovations for which pure statistical models are not appropriate in principle. This work proposes a memory augmented neural model for Chinese poem generation, where the neural model and the augmented memory work together to balance the requirements of linguistic accordance and aesthetic innovation, leading to innovative generations that are still rule-compliant. In addition, it is found that the memory mechanism provides interesting flexibility that can be used to generate poems with different styles.