ACL 2019 论文列表

Unsupervised bilingual word embedding (UBWE), together with other technologies such as back-translation and denoising, has helped unsupervised neural machine translation (UNMT) achieve remarkable results in several language pairs. In previous methods, UBWE is first trained using non-parallel monolingual corpora and then this pre-trained UBWE is used to initialize the word embedding in the encoder and decoder of UNMT. That is, the training of UBWE and UNMT are separate. In this paper, we first empirically investigate the relationship between UBWE and UNMT. The empirical findings show that the performance of UNMT is significantly affected by the performance of UBWE. Thus, we propose two methods that train UNMT with UBWE agreement. Empirical results on several language pairs show that the proposed methods significantly outperform conventional UNMT.

Mining parallel sentences from comparable corpora is important. Most previous work relies on supervised systems, which are trained on parallel data, thus their applicability is problematic in low-resource scenarios. Recent developments in building unsupervised bilingual word embeddings made it possible to mine parallel sentences based on cosine similarities of source and target language words. We show that relying only on this information is not enough, since sentences often have similar words but different meanings. We detect continuous parallel segments in sentence pair candidates and rely on them when mining parallel sentences. We show better mining accuracy on three language pairs in a standard shared task on artificial data. We also provide the first experiments showing that parallel sentences mined from real life sources improve unsupervised MT. Our code is available, we hope it will be used to support low-resource MT research.

Multilingual neural machine translation (Multi-NMT) with one encoder-decoder model has made remarkable progress due to its simple deployment. However, this multilingual translation paradigm does not make full use of language commonality and parameter sharing between encoder and decoder. Furthermore, this kind of paradigm cannot outperform the individual models trained on bilingual corpus in most cases. In this paper, we propose a compact and language-sensitive method for multilingual translation. To maximize parameter sharing, we first present a universal representor to replace both encoder and decoder models. To make the representor sensitive for specific languages, we further introduce language-sensitive embedding, attention, and discriminator with the ability to enhance model performance. We verify our methods on various translation scenarios, including one-to-many, many-to-many and zero-shot. Extensive experiments demonstrate that our proposed methods remarkably outperform strong standard multilingual translation systems on WMT and IWSLT datasets. Moreover, we find that our model is especially helpful in low-resource and zero-shot translation scenarios.

Though machine translation errors caused by the lack of context beyond one sentence have long been acknowledged, the development of context-aware NMT systems is hampered by several problems. Firstly, standard metrics are not sensitive to improvements in consistency in document-level translations. Secondly, previous work on context-aware NMT assumed that the sentence-aligned parallel data consisted of complete documents while in most practical scenarios such document-level data constitutes only a fraction of the available parallel data. To address the first issue, we perform a human study on an English-Russian subtitles dataset and identify deixis, ellipsis and lexical cohesion as three main sources of inconsistency. We then create test sets targeting these phenomena. To address the second shortcoming, we consider a set-up in which a much larger amount of sentence-level data is available compared to that aligned at the document level. We introduce a model that is suitable for this scenario and demonstrate major gains over a context-agnostic baseline on our new benchmarks without sacrificing performance as measured with BLEU.

Lattices are an efficient and effective method to encode ambiguity of upstream systems in natural language processing tasks, for example to compactly capture multiple speech recognition hypotheses, or to represent multiple linguistic analyses. Previous work has extended recurrent neural networks to model lattice inputs and achieved improvements in various tasks, but these models suffer from very slow computation speeds. This paper extends the recently proposed paradigm of self-attention to handle lattice inputs. Self-attention is a sequence modeling technique that relates inputs to one another by computing pairwise similarities and has gained popularity for both its strong results and its computational efficiency. To extend such models to handle lattices, we introduce probabilistic reachability masks that incorporate lattice structure into the model and support lattice scores if available. We also propose a method for adapting positional embeddings to lattice structures. We apply the proposed model to a speech translation task and find that it outperforms all examined baselines while being much faster to compute than previous neural lattice models during both training and inference.

Human trafficking is a worldwide crisis. Traffickers exploit their victims by anonymously offering sexual services through online advertisements. These ads often contain clues that law enforcement can use to separate out potential trafficking cases from volunteer sex advertisements. The problem is that the sheer volume of ads is too overwhelming for manual processing. Ideally, a centralized semi-automated tool can be used to assist law enforcement agencies with this task. Here, we present an approach using natural language processing to identify trafficking ads on these websites. We propose a classifier by integrating multiple text feature sets, including the publicly available pre-trained textual language model Bi-directional Encoder Representation from transformers (BERT). In this paper, we demonstrate that a classifier using this composite feature set has significantly better performance compared to any single feature set alone.

In this study, we propose a new multi-task learning approach for rumor detection and stance classification tasks. This neural network model has a shared layer and two task specific layers. We incorporate the user credibility information into the rumor detection layer, and we also apply attention mechanism in the rumor detection process. The attended information include not only the hidden states in the rumor detection layer, but also the hidden states from the stance detection layer. The experiments on two datasets show that our proposed model outperforms the state-of-the-art rumor detection approaches.

In visual communication, text emphasis is used to increase the comprehension of written text to convey the author’s intent. We study the problem of emphasis selection, i.e. choosing candidates for emphasis in short written text, to enable automated design assistance in authoring. Without knowing the author’s intent and only considering the input text, multiple emphasis selections are valid. We propose a model that employs end-to-end label distribution learning (LDL) on crowd-sourced data and predicts a selection distribution, capturing the inter-subjectivity (common-sense) in the audience as well as the ambiguity of the input. We compare the model with several baselines in which the problem is transformed to single-label learning by mapping label distributions to absolute labels via majority voting.

The word ordering in a Sanskrit verse is often not aligned with its corresponding prose order. Conversion of the verse to its corresponding prose helps in better comprehension of the construction. Owing to the resource constraints, we formulate this task as a word ordering (linearisation) task. In doing so, we completely ignore the word arrangement at the verse side. kāvya guru, the approach we propose, essentially consists of a pipeline of two pretraining steps followed by a seq2seq model. The first pretraining step learns task-specific token embeddings from pretrained embeddings. In the next step, we generate multiple possible hypotheses for possible word arrangements of the input %using another pretraining step. We then use them as inputs to a neural seq2seq model for the final prediction. We empirically show that the hypotheses generated by our pretraining step result in predictions that consistently outperform predictions based on the original order in the verse. Overall, kāvya guru outperforms current state of the art models in linearisation for the poetry to prose conversion task in Sanskrit.

News recommendation can help users find interested news and alleviate information overload. The topic information of news is critical for learning accurate news and user representations for news recommendation. However, it is not considered in many existing news recommendation methods. In this paper, we propose a neural news recommendation approach with topic-aware news representations. The core of our approach is a topic-aware news encoder and a user encoder. In the news encoder we learn representations of news from their titles via CNN networks and apply attention networks to select important words. In addition, we propose to learn topic-aware news representations by jointly training the news encoder with an auxiliary topic classification task. In the user encoder we learn the representations of users from their browsed news and use attention networks to select informative news for user representation learning. Extensive experiments on a real-world dataset validate the effectiveness of our approach.

Complex Word Identification (CWI) is concerned with detection of words in need of simplification and is a crucial first step in a simplification pipeline. It has been shown that reliable CWI systems considerably improve text simplification. However, most CWI systems to date address the task on a word-by-word basis, not taking the context into account. In this paper, we present a novel approach to CWI based on sequence modelling. Our system is capable of performing CWI in context, does not require extensive feature engineering and outperforms state-of-the-art systems on this task.

Personal health mention detection deals with predicting whether or not a given sentence is a report of a health condition. Past work mentions errors in this prediction when symptom words, i.e., names of symptoms of interest, are used in a figurative sense. Therefore, we combine a state-of-the-art figurative usage detection with CNN-based personal health mention detection. To do so, we present two methods: a pipeline-based approach and a feature augmentation-based approach. The introduction of figurative usage detection results in an average improvement of 2.21% F-score of personal health mention detection, in the case of the feature augmentation-based approach. This paper demonstrates the promise of using figurative usage detection to improve personal health mention detection.

We present a novel conversational-context aware end-to-end speech recognizer based on a gated neural network that incorporates conversational-context/word/speech embeddings. Unlike conventional speech recognition models, our model learns longer conversational-context information that spans across sentences and is consequently better at recognizing long conversations. Specifically, we propose to use text-based external word and/or sentence embeddings (i.e., fastText, BERT) within an end-to-end framework, yielding significant improvement in word error rate with better conversational-context representation. We evaluated the models on the Switchboard conversational speech corpus and show that our model outperforms standard end-to-end speech recognition models.

Automatically validating a research artefact is one of the frontiers in Artificial Intelligence (AI) that directly brings it close to competing with human intellect and intuition. Although criticised sometimes, the existing peer review system still stands as the benchmark of research validation. The present-day peer review process is not straightforward and demands profound domain knowledge, expertise, and intelligence of human reviewer(s), which is somewhat elusive with the current state of AI. However, the peer review texts, which contains rich sentiment information of the reviewer, reflecting his/her overall attitude towards the research in the paper, could be a valuable entity to predict the acceptance or rejection of the manuscript under consideration. Here in this work, we investigate the role of reviewer sentiment embedded within peer review texts to predict the peer review outcome. Our proposed deep neural architecture takes into account three channels of information: the paper, the corresponding reviews, and review’s polarity to predict the overall recommendation score as well as the final decision. We achieve significant performance improvement over the baselines (∼ 29% error reduction) proposed in a recently released dataset of peer reviews. An AI of this kind could assist the editors/program chairs as an additional layer of confidence, especially when non-responding/missing reviewers are frequent in present day peer review.

Distributions of the senses of words are often highly skewed and give a strong influence of the domain of a document. This paper follows the assumption and presents a method for text categorization by leveraging the predominant sense of words depending on the domain, i.e., domain-specific senses. The key idea is that the features learned from predominant senses are possible to discriminate the domain of the document and thus improve the overall performance of text categorization. We propose multi-task learning framework based on the neural network model, transformer, which trains a model to simultaneously categorize documents and predicts a predominant sense for each word. The experimental results using four benchmark datasets show that our method is comparable to the state-of-the-art categorization approach, especially our model works well for categorization of multi-label documents.

Authorship verification is the task of determining whether two texts were written by the same author. We deal with the adversary task, called authorship obfuscation: preventing verification by altering a to-be-obfuscated text. Our new obfuscation approach (1) models writing style difference as the Jensen-Shannon distance between the character n-gram distributions of texts, and (2) manipulates an author’s subconsciously encoded writing style in a sophisticated manner using heuristic search. To obfuscate, we analyze the huge space of textual variants for a paraphrased version of the to-be-obfuscated text that has a sufficient Jensen-Shannon distance at minimal costs in terms of text quality. We analyze, quantify, and illustrate the rationale of this approach, define paraphrasing operators, derive obfuscation thresholds, and develop an effective obfuscation framework. Our authorship obfuscation approach defeats state-of-the-art verification approaches, including unmasking and compression models, while keeping text changes at a minimum.

We address the problem of adversarial attacks on text classification, which is rarely studied comparing to attacks on image classification. The challenge of this task is to generate adversarial examples that maintain lexical correctness, grammatical correctness and semantic similarity. Based on the synonyms substitution strategy, we introduce a new word replacement order determined by both the word saliency and the classification probability, and propose a greedy algorithm called probability weighted word saliency (PWWS) for text adversarial attack. Experiments on three popular datasets using convolutional as well as LSTM models show that PWWS reduces the classification accuracy to the most extent, and keeps a very low word substitution rate. A human evaluation study shows that our generated adversarial examples maintain the semantic similarity well and are hard for humans to perceive. Performing adversarial training using our perturbed datasets improves the robustness of the models. At last, our method also exhibits a good transferability on the generated adversarial examples.

Automatic generation of summaries from multiple news articles is a valuable tool as the number of online publications grows rapidly. Single document summarization (SDS) systems have benefited from advances in neural encoder-decoder model thanks to the availability of large datasets. However, multi-document summarization (MDS) of news articles has been limited to datasets of a couple of hundred examples. In this paper, we introduce Multi-News, the first large-scale MDS news dataset. Additionally, we propose an end-to-end model which incorporates a traditional extractive summarization model with a standard SDS model and achieves competitive results on MDS datasets. We benchmark several methods on Multi-News and hope that this work will promote advances in summarization in the multi-document setting.

Research on summarization has mainly been driven by empirical approaches, crafting systems to perform well on standard datasets with the notion of information Importance remaining latent. We argue that establishing theoretical models of Importance will advance our understanding of the task and help to further improve summarization systems. To this end, we propose simple but rigorous definitions of several concepts that were previously used only intuitively in summarization: Redundancy, Relevance, and Informativeness. Importance arises as a single quantity naturally unifying these concepts. Additionally, we provide intuitions to interpret the proposed quantities and experiments to demonstrate the potential of the framework to inform and guide subsequent works.