NAACL 2019 论文列表

We introduce a novel method for multilingual transfer that utilizes deep contextual embeddings, pretrained in an unsupervised fashion. While contextual embeddings have been shown to yield richer representations of meaning compared to their static counterparts, aligning them poses a challenge due to their dynamic nature. To this end, we construct context-independent variants of the original monolingual spaces and utilize their mapping to derive an alignment for the context-dependent spaces. This mapping readily supports processing of a target language, improving transfer by context-aware embeddings. Our experimental results demonstrate the effectiveness of this approach for zero-shot and few-shot learning of dependency parsing. Specifically, our method consistently outperforms the previous state-of-the-art on 6 tested languages, yielding an improvement of 6.8 LAS points on average.

Recent approaches to cross-lingual word embedding have generally been based on linear transformations between the sets of embedding vectors in the two languages. In this paper, we propose an approach that instead expresses the two monolingual embedding spaces as probability densities defined by a Gaussian mixture model, and matches the two densities using a method called normalizing flow. The method requires no explicit supervision, and can be learned with only a seed dictionary of words that have identical strings. We argue that this formulation has several intuitively attractive properties, particularly with the respect to improving robustness and generalization to mappings between difficult language pairs or word pairs. On a benchmark data set of bilingual lexicon induction and cross-lingual word similarity, our approach can achieve competitive or superior performance compared to state-of-the-art published results, with particularly strong results being found on etymologically distant and/or morphologically rich languages.

Combinatory categorial grammars are linguistically motivated and useful for semantic parsing, but costly to acquire in a supervised way and difficult to acquire in an unsupervised way. We propose an alternative making use of cross-lingual learning: an existing source-language parser is used together with a parallel corpus to induce a grammar and parsing model for a target language. On the PASCAL benchmark, cross-lingual CCG induction outperforms CCG induction from gold-standard POS tags on 3 out of 8 languages, and unsupervised CCG induction on 6 out of 8 languages. We also show that cross-lingually induced CCGs reflect syntactic properties of the target languages.

The need for tree structure modelling on top of sequence modelling is an open issue in neural dependency parsing. We investigate the impact of adding a tree layer on top of a sequential model by recursively composing subtree representations (composition) in a transition-based parser that uses features extracted by a BiLSTM. Composition seems superfluous with such a model, suggesting that BiLSTMs capture information about subtrees. We perform model ablations to tease out the conditions under which composition helps. When ablating the backward LSTM, performance drops and composition does not recover much of the gap. When ablating the forward LSTM, performance drops less dramatically and composition recovers a substantial part of the gap, indicating that a forward LSTM and composition capture similar information. We take the backward LSTM to be related to lookahead features and the forward LSTM to the rich history-based features both crucial for transition-based parsers. To capture history-based information, composition is better than a forward LSTM on its own, but it is even better to have a forward LSTM as part of a BiLSTM. We correlate results with language properties, showing that the improved lookahead of a backward LSTM is especially important for head-final languages.

We introduce a theoretical analysis of crosslingual transfer in probabilistic topic models. By formulating posterior inference through Gibbs sampling as a process of language transfer, we propose a new measure that quantifies the loss of knowledge across languages during this process. This measure enables us to derive a PAC-Bayesian bound that elucidates the factors affecting model quality, both during training and in downstream applications. We provide experimental validation of the analysis on a diverse set of five languages, and discuss best practices for data collection and model design based on our analysis.

Brown and Exchange word clusters have long been successfully used as word representations in Natural Language Processing (NLP) systems. Their success has been attributed to their seeming ability to represent both semantic and syntactic information. Using corpora representing several language families, we test the hypothesis that Brown and Exchange word clusters are highly effective at encoding morphosyntactic information. Our experiments show that word clusters are highly capable at distinguishing Parts of Speech. We show that increases in Average Mutual Information, the clustering algorithms’ optimization goal, are highly correlated with improvements in encoding of morphosyntactic information. Our results provide empirical evidence that downstream NLP systems addressing tasks dependent on morphosyntactic information can benefit from word cluster features.

In the principles-and-parameters framework, the structural features of languages depend on parameters that may be toggled on or off, with a single parameter often dictating the status of multiple features. The implied covariance between features inspires our probabilisation of this line of linguistic inquiry—we develop a generative model of language based on exponential-family matrix factorisation. By modelling all languages and features within the same architecture, we show how structural similarities between languages can be exploited to predict typological features with near-perfect accuracy, outperforming several baselines on the task of predicting held-out features. Furthermore, we show that language embeddings pre-trained on monolingual text allow for generalisation to unobserved languages. This finding has clear practical and also theoretical implications: the results confirm what linguists have hypothesised, i.e. that there are significant correlations between typological features and languages.

English verbs have multiple forms. For instance, talk may also appear as talks, talked or talking, depending on the context. The NLP task of lemmatization seeks to map these diverse forms back to a canonical one, known as the lemma. We present a simple joint neural model for lemmatization and morphological tagging that achieves state-of-the-art results on 20 languages from the Universal Dependencies corpora. Our paper describes the model in addition to training and decoding procedures. Error analysis indicates that joint morphological tagging and lemmatization is especially helpful in low-resource lemmatization and languages that display a larger degree of morphological complexity.

Recent work has shown that contextualized word representations derived from neural machine translation are a viable alternative to such from simple word predictions tasks. This is because the internal understanding that needs to be built in order to be able to translate from one language to another is much more comprehensive. Unfortunately, computational and memory limitations as of present prevent NMT models from using large word vocabularies, and thus alternatives such as subword units (BPE and morphological segmentations) and characters have been used. Here we study the impact of using different kinds of units on the quality of the resulting representations when used to model morphology, syntax, and semantics. We found that while representations derived from subwords are slightly better for modeling syntax, character-based representations are superior for modeling morphology and are also more robust to noisy input.

Lemmatization of standard languages is concerned with (i) abstracting over morphological differences and (ii) resolving token-lemma ambiguities of inflected words in order to map them to a dictionary headword. In the present paper we aim to improve lemmatization performance on a set of non-standard historical languages in which the difficulty is increased by an additional aspect (iii): spelling variation due to lacking orthographic standards. We approach lemmatization as a string-transduction task with an Encoder-Decoder architecture which we enrich with sentence information using a hierarchical sentence encoder. We show significant improvements over the state-of-the-art by fine-tuning the sentence encodings to jointly optimize a bidirectional language model loss. Crucially, our architecture does not require POS or morphological annotations, which are not always available for historical corpora. Additionally, we also test the proposed model on a set of typologically diverse standard languages showing results on par or better than a model without fine-tuned sentence representations and previous state-of-the-art systems. Finally, to encourage future work on processing of non-standard varieties, we release the dataset of non-standard languages underlying the present study, which is based on openly accessible sources.

Human phenotype-gene relations are fundamental to fully understand the origin of some phenotypic abnormalities and their associated diseases. Biomedical literature is the most comprehensive source of these relations, however, we need Relation Extraction tools to automatically recognize them. Most of these tools require an annotated corpus and to the best of our knowledge, there is no corpus available annotated with human phenotype-gene relations. This paper presents the Phenotype-Gene Relations (PGR) corpus, a silver standard corpus of human phenotype and gene annotations and their relations. The corpus consists of 1712 abstracts, 5676 human phenotype annotations, 13835 gene annotations, and 4283 relations. We generated this corpus using Named-Entity Recognition tools, whose results were partially evaluated by eight curators, obtaining a precision of 87.01%. By using the corpus we were able to obtain promising results with two state-of-the-art deep learning tools, namely 78.05% of precision. The PGR corpus was made publicly available to the research community.

Nowadays social media platforms are the most popular way for people to share information, from work issues to personal matters. For example, people with health disorders tend to share their concerns for advice, support or simply to relieve suffering. This provides a great opportunity to proactively detect these users and refer them as soon as possible to professional help. We propose a new representation called Bag of Sub-Emotions (BoSE), which represents social media documents by a set of fine-grained emotions automatically generated using a lexical resource of emotions and subword embeddings. The proposed representation is evaluated in the task of depression detection. The results are encouraging; the usage of fine-grained emotions improved the results from a representation based on the core emotions and obtained competitive results in comparison to state of the art approaches.

Modern NLP systems require high-quality annotated data. For specialized domains, expert annotations may be prohibitively expensive; the alternative is to rely on crowdsourcing to reduce costs at the risk of introducing noise. In this paper we demonstrate that directly modeling instance difficulty can be used to improve model performance and to route instances to appropriate annotators. Our difficulty prediction model combines two learned representations: a ‘universal’ encoder trained on out of domain data, and a task-specific encoder. Experiments on a complex biomedical information extraction task using expert and lay annotators show that: (i) simply excluding from the training data instances predicted to be difficult yields a small boost in performance; (ii) using difficulty scores to weight instances during training provides further, consistent gains; (iii) assigning instances predicted to be difficult to domain experts is an effective strategy for task routing. Further, our experiments confirm the expectation that for such domain-specific tasks expert annotations are of much higher quality and preferable to obtain if practical and that augmenting small amounts of expert data with a larger set of lay annotations leads to further improvements in model performance.

Word vectors and Language Models (LMs) pretrained on a large amount of unlabelled data can dramatically improve various Natural Language Processing (NLP) tasks. However, the measure and impact of similarity between pretraining data and target task data are left to intuition. We propose three cost-effective measures to quantify different aspects of similarity between source pretraining and target task data. We demonstrate that these measures are good predictors of the usefulness of pretrained models for Named Entity Recognition (NER) over 30 data pairs. Results also suggest that pretrained LMs are more effective and more predictable than pretrained word vectors, but pretrained word vectors are better when pretraining data is dissimilar.

The recent surge of text-based online counseling applications enables us to collect and analyze interactions between counselors and clients. A dataset of those interactions can be used to learn to automatically classify the client utterances into categories that help counselors in diagnosing client status and predicting counseling outcome. With proper anonymization, we collect counselor-client dialogues, define meaningful categories of client utterances with professional counselors, and develop a novel neural network model for classifying the client utterances. The central idea of our model, ConvMFiT, is a pre-trained conversation model which consists of a general language model built from an out-of-domain corpus and two role-specific language models built from unlabeled in-domain dialogues. The classification result shows that ConvMFiT outperforms state-of-the-art comparison models. Further, the attention weights in the learned model confirm that the model finds expected linguistic patterns for each category.

Relation extraction (RE) aims to label relations between groups of marked entities in raw text. Most current RE models learn context-aware representations of the target entities that are then used to establish relation between them. This works well for intra-sentence RE, and we call them first-order relations. However, this methodology can sometimes fail to capture complex and long dependencies. To address this, we hypothesize that at times the target entities can be connected via a context token. We refer to such indirect relations as second-order relations, and describe an efficient implementation for computing them. These second-order relation scores are then combined with first-order relation scores to obtain final relation scores. Our empirical results show that the proposed method leads to state-of-the-art performance over two biomedical datasets.

Linguistic features have shown promising applications for detecting various cognitive impairments. To improve detection accuracies, increasing the amount of data or the number of linguistic features have been two applicable approaches. However, acquiring additional clinical data can be expensive, and hand-crafting features is burdensome. In this paper, we take a third approach, proposing Consensus Networks (CNs), a framework to classify after reaching agreements between modalities. We divide linguistic features into non-overlapping subsets according to their modalities, and let neural networks learn low-dimensional representations that agree with each other. These representations are passed into a classifier network. All neural networks are optimized iteratively. In this paper, we also present two methods that improve the performance of CNs. We then present ablation studies to illustrate the effectiveness of modality division. To understand further what happens in CNs, we visualize the representations during training. Overall, using all of the 413 linguistic features, our models significantly outperform traditional classifiers, which are used by the state-of-the-art papers.

Event extraction for the biomedical domain is more challenging than that in the general news domain since it requires broader acquisition of domain-specific knowledge and deeper understanding of complex contexts. To better encode contextual information and external background knowledge, we propose a novel knowledge base (KB)-driven tree-structured long short-term memory networks (Tree-LSTM) framework, incorporating two new types of features: (1) dependency structures to capture wide contexts; (2) entity properties (types and category descriptions) from external ontologies via entity linking. We evaluate our approach on the BioNLP shared task with Genia dataset and achieve a new state-of-the-art result. In addition, both quantitative and qualitative studies demonstrate the advancement of the Tree-LSTM and the external knowledge representation for biomedical event extraction.

As offensive content has become pervasive in social media, there has been much research in identifying potentially offensive messages. However, previous work on this topic did not consider the problem as a whole, but rather focused on detecting very specific types of offensive content, e.g., hate speech, cyberbulling, or cyber-aggression. In contrast, here we target several different kinds of offensive content. In particular, we model the task hierarchically, identifying the type and the target of offensive messages in social media. For this purpose, we complied the Offensive Language Identification Dataset (OLID), a new dataset with tweets annotated for offensive content using a fine-grained three-layer annotation scheme, which we make publicly available. We discuss the main similarities and differences between OLID and pre-existing datasets for hate speech identification, aggression detection, and similar tasks. We further experiment with and we compare the performance of different machine learning models on OLID.