Topic models are a popular tool for understanding text collections, but their evaluation has been a point of contention. Automated evaluation metrics such as coherence are often used, however, their validity has been questioned for neural topic models (NTMs) and can overlook a model’s benefits in real-world applications. To this end, we conduct the first evaluation of neural, supervised and classical topic models in an interactive task-based setting. We combine topic models with a classifier and test their ability to help humans conduct content analysis and document annotation. From simulated, real user and expert pilot studies, the Contextual Neural Topic Model does the best on cluster evaluation metrics and human evaluations; however, LDA is competitive with two other NTMs under our simulated experiment and user study results, contrary to what coherence scores suggest. We show that current automated metrics do not provide a complete picture of topic modeling capabilities, but the right choice of NTMs can be better than classical models on practical tasks.

When people interpret text, they rely on inferences that go beyond the observed language itself. Inspired by this observation, we introduce a method for the analysis of text that takes implicitly communicated content explicitly into account. We use a large language model to produce sets of propositions that are inferentially related to the text that has been observed, then validate the plausibility of the generated content via human judgments. Incorporating these explicit representations of implicit content proves useful in multiple problem settings that involve the human interpretation of utterances: assessing the similarity of arguments, making sense of a body of opinion data, and modeling legislative behavior. Our results suggest that modeling the meanings behind observed language, rather than the literal text alone, is a valuable direction for NLP and particularly its applications to social science.

Recently, the relationship between automated and human evaluation of topic models has been called into question. Method developers have staked the efficacy of new topic model variants on automated measures, and their failure to approximate human preferences places these models on uncertain ground. Moreover, existing evaluation paradigms are often divorced from real-world use.Motivated by content analysis as a dominant real-world use case for topic modeling, we analyze two related aspects of topic models that affect their effectiveness and trustworthiness in practice for that purpose: the stability of their estimates and the extent to which the model’s discovered categories align with human-determined categories in the data. We find that neural topic models fare worse in both respects compared to an established classical method. We take a step toward addressing both issues in tandem by demonstrating that a straightforward ensembling method can reliably outperform the members of the ensemble.

Human innovation in language, such as inventing new words, is a challenge for pretrained language models. We assess the ability of one large model, GPT-3, to process new words and decide on their meaning. We create a set of nonce words and prompt GPT-3 to generate their dictionary definitions. We find GPT-3 produces plausible definitions that align with human judgments. Moreover, GPT-3’s definitions are sometimes preferred to those invented by humans, signaling its intriguing ability not just to adapt, but to add to the evolving vocabulary of the English language.

As neural language models approach human performance on NLP benchmark tasks, their advances are widely seen as evidence of an increasingly complex understanding of syntax. This view rests upon a hypothesis that has not yet been empirically tested: that word order encodes meaning essential to performing these tasks. We refute this hypothesis in many cases: in the GLUE suite and in various genres of English text, the words in a sentence or phrase can rarely be permuted to form a phrase carrying substantially different information. Our surprising result relies on inference by iterative shuffling (IBIS), a novel, efficient procedure that finds the ordering of a bag of words having the highest likelihood under a fixed language model. IBIS can use any black-box model without additional training and is superior to existing word ordering algorithms. Coalescing our findings, we discuss how shuffling inference procedures such as IBIS can benefit language modeling and constrained generation.

ISO-TimeML is an international standard for multilingual event annotation, detection, categorization and linking. In this paper, we present the Hindi TimeBank, an ISO-TimeML annotated reference corpus for the detection and classification of events, states and time expressions, and the links between them. Based on contemporary developments in Hindi event recognition, we propose language independent and language-specific deviations from the ISO-TimeML guidelines, but preserve the schema. These deviations include the inclusion of annotator confidence, and an independent mechanism of identifying and annotating states such as copulars and existentials) With this paper, we present an open-source corpus, the Hindi TimeBank. The Hindi TimeBank is a 1,000 article dataset, with over 25,000 events, 3,500 states and 2,000 time expressions. We analyze the dataset in detail and provide a class-wise distribution of events, states and time expressions. Our guidelines and dataset are backed by high average inter-annotator agreement scores.

Topic models are often used to identify human-interpretable topics to help make sense of large document collections. We use knowledge distillation to combine the best attributes of probabilistic topic models and pretrained transformers. Our modular method can be straightforwardly applied with any neural topic model to improve topic quality, which we demonstrate using two models having disparate architectures, obtaining state-of-the-art topic coherence. We show that our adaptable framework not only improves performance in the aggregate over all estimated topics, as is commonly reported, but also in head-to-head comparisons of aligned topics.

Detection of TimeML events in text have traditionally been done on corpora such as TimeBanks. However, deep learning methods have not been applied to these corpora, because these datasets seldom contain more than 10,000 event mentions. Traditional architectures revolve around highly feature engineered, language specific statistical models. In this paper, we present a Language Invariant Neural Event Detection (ALINED) architecture. ALINED uses an aggregation of both sub-word level features as well as lexical and structural information. This is achieved by combining convolution over character embeddings, with recurrent layers over contextual word embeddings. We find that our model extracts relevant features for event span identification without relying on language specific features. We compare the performance of our language invariant model to the current state-of-the-art in English, Spanish, Italian and French. We outperform the F1-score of the state of the art in English by 1.65 points. We achieve F1-scores of 84.96, 80.87 and 74.81 on Spanish, Italian and French respectively which is comparable to the current states of the art for these languages. We also introduce the automatic annotation of events in Hindi, a low resource language, with an F1-Score of 77.13.

We describe the development of a knowledge graph from an event annotated corpus by presenting a pipeline that identifies and extracts the relations between entities and events from Hindi news articles. Due to the semantic implications of argument identification for events in Hindi, we use a combined syntactic argument and semantic role identification methodology. To the best of our knowledge, no other architecture exists for this purpose. The extracted combined role information is incorporated in a knowledge graph that can be queried via subgraph extraction for basic questions. The architectures presented in this paper can be used for participant extraction and event-entity linking in most Indo-Aryan languages, due to similar syntactic and semantic properties of event arguments.

Understanding common sense is important for effective natural language reasoning. One type of common sense is how two objects compare on physical properties such as size and weight: e.g., ‘is a house bigger than a person?’. We probe whether pre-trained representations capture comparisons and find they, in fact, have higher accuracy than previous approaches. They also generalize to comparisons involving objects not seen during training. We investigate how such comparisons are made: models learn a consistent ordering over all the objects in the comparisons. Probing models have significantly higher accuracy than those baseline models which use dataset artifacts: e.g., memorizing some words are larger than any other word.

Social media based micro-blogging sites like Twitter have become a common source of real-time information (impacting organizations and their strategies, and are used for expressing emotions and opinions. Automated analysis of such content therefore rises in importance. To this end, we explore the viability of using deep neural networks on the specific task of emotion intensity prediction in tweets. We propose a neural architecture combining convolutional and fully connected layers in a non-sequential manner - done for the first time in context of natural language based tasks. Combined with lexicon-based features along with transfer learning, our model achieves state-of-the-art performance, outperforming the previous system by 0.044 or 4.4% Pearson correlation on the WASSA’17 EmoInt shared task dataset. We investigate the performance of deep multi-task learning models trained for all emotions at once in a unified architecture and get encouraging results. Experiments performed on evaluating correlation between emotion pairs offer interesting insights into the relationship between them.

This paper describes an ensemble system submitted as part of the LSDSem Shared Task 2017 - the Story Cloze Test. The main conclusion from our results is that an approach based on semantic similarity alone may not be enough for this task. We test various approaches and compare them with two ensemble systems. One is based on voting and the other on logistic regression based classifier. Our final system is able to outperform the previous state of the art for the Story Cloze test. Another very interesting observation is the performance of sentiment based approach which works almost as well on its own as our final ensemble system.

The paper describes the best performing system for EmoInt - a shared task to predict the intensity of emotions in tweets. Intensity is a real valued score, between 0 and 1. The emotions are classified as - anger, fear, joy and sadness. We apply three different deep neural network based models, which approach the problem from essentially different directions. Our final performance quantified by an average pearson correlation score of 74.7 and an average spearman correlation score of 73.5 is obtained using an ensemble of the three models. We outperform the baseline model of the shared task by 9.9% and 9.4% pearson and spearman correlation scores respectively.