ACL 2019 论文列表

Motivated by infamous cheating scandals in the media industry, the wine industry, and political campaigns, we address the problem of detecting concealed information in technical settings. In this work, we explore acoustic-prosodic and linguistic indicators of information concealment by collecting a unique corpus of professionals practicing for oral exams while concealing information. We reveal subtle signs of concealing information in speech and text, compare and contrast them with those in deception detection literature, uncovering the link between concealing information and deception. We then present a series of experiments that automatically detect concealed information from text and speech. We compare the use of acoustic-prosodic, linguistic, and individual feature sets, using different machine learning models. Finally, we present a multi-task learning framework with acoustic, linguistic, and individual features, that outperforms human performance by over 15%.

Predicting financial risk is an essential task in financial market. Prior research has shown that textual information in a firm’s financial statement can be used to predict its stock’s risk level. Nowadays, firm CEOs communicate information not only verbally through press releases and financial reports, but also nonverbally through investor meetings and earnings conference calls. There are anecdotal evidences that CEO’s vocal features, such as emotions and voice tones, can reveal the firm’s performance. However, how vocal features can be used to predict risk levels, and to what extent, is still unknown. To fill the gap, we obtain earnings call audio recordings and textual transcripts for S&P 500 companies in recent years. We propose a multimodal deep regression model (MDRM) that jointly model CEO’s verbal (from text) and vocal (from audio) information in a conference call. Empirical results show that our model that jointly considers verbal and vocal features achieves significant and substantial prediction error reduction. We also discuss several interesting findings and the implications to financial markets. The processed earnings conference calls data (text and audio) are released for readers who are interested in reproducing the results or designing trading strategy.

Clinical letters are infamously impenetrable for the lay patient. This work uses neural text simplification methods to automatically improve the understandability of clinical letters for patients. We take existing neural text simplification software and augment it with a new phrase table that links complex medical terminology to simpler vocabulary by mining SNOMED-CT. In an evaluation task using crowdsourcing, we show that the results of our new system are ranked easier to understand (average rank 1.93) than using the original system (2.34) without our phrase table. We also show improvement against baselines including the original text (2.79) and using the phrase table without the neural text simplification software (2.94). Our methods can easily be transferred outside of the clinical domain by using domain-appropriate resources to provide effective neural text simplification for any domain without the need for costly annotation.

Text classification aims at mapping documents into a set of predefined categories. Supervised machine learning models have shown great success in this area but they require a large number of labeled documents to reach adequate accuracy. This is particularly true when the number of target categories is in the tens or the hundreds. In this work, we explore an unsupervised approach to classify documents into categories simply described by a label. The proposed method is inspired by the way a human proceeds in this situation: It draws on textual similarity between the most relevant words in each document and a dictionary of keywords for each category reflecting its semantics and lexical field. The novelty of our method hinges on the enrichment of the category labels through a combination of human expertise and language models, both generic and domain specific. Our experiments on 5 standard corpora show that the proposed method increases F1-score over relying solely on human expertise and can also be on par with simple supervised approaches. It thus provides a practical alternative to situations where low cost text categorization is needed, as we illustrate with our application to operational risk incidents classification.

We propose two novel manipulation strategies for increasing and decreasing the difficulty of C-tests automatically. This is a crucial step towards generating learner-adaptive exercises for self-directed language learning and preparing language assessment tests. To reach the desired difficulty level, we manipulate the size and the distribution of gaps based on absolute and relative gap difficulty predictions. We evaluate our approach in corpus-based experiments and in a user study with 60 participants. We find that both strategies are able to generate C-tests with the desired difficulty level.

In this paper, we automatically create sentiment dictionaries for predicting financial outcomes. We compare three approaches: (i) manual adaptation of the domain-general dictionary H4N, (ii) automatic adaptation of H4N and (iii) a combination consisting of first manual, then automatic adaptation. In our experiments, we demonstrate that the automatically adapted sentiment dictionary outperforms the previous state of the art in predicting the financial outcomes excess return and volatility. In particular, automatic adaptation performs better than manual adaptation. In our analysis, we find that annotation based on an expert’s a priori belief about a word’s meaning can be incorrect – annotation should be performed based on the word’s contexts in the target domain instead.

Personalized news recommendation is important to help users find their interested news and improve reading experience. A key problem in news recommendation is learning accurate user representations to capture their interests. Users usually have both long-term preferences and short-term interests. However, existing news recommendation methods usually learn single representations of users, which may be insufficient. In this paper, we propose a neural news recommendation approach which can learn both long- and short-term user representations. The core of our approach is a news encoder and a user encoder. In the news encoder, we learn representations of news from their titles and topic categories, and use attention network to select important words. In the user encoder, we propose to learn long-term user representations from the embeddings of their IDs.In addition, we propose to learn short-term user representations from their recently browsed news via GRU network. Besides, we propose two methods to combine long-term and short-term user representations. The first one is using the long-term user representation to initialize the hidden state of the GRU network in short-term user representation. The second one is concatenating both long- and short-term user representations as a unified user vector. Extensive experiments on a real-world dataset show our approach can effectively improve the performance of neural news recommendation.

We propose a novel self-attention mechanism that can learn its optimal attention span. This allows us to extend significantly the maximum context size used in Transformer, while maintaining control over their memory footprint and computational time. We show the effectiveness of our approach on the task of character level language modeling, where we achieve state-of-the-art performances on text8 and enwiki8 by using a maximum context of 8k characters.

Neural parsers obtain state-of-the-art results on benchmark treebanks for constituency parsing—but to what degree do they generalize to other domains? We present three results about the generalization of neural parsers in a zero-shot setting: training on trees from one corpus and evaluating on out-of-domain corpora. First, neural and non-neural parsers generalize comparably to new domains. Second, incorporating pre-trained encoder representations into neural parsers substantially improves their performance across all domains, but does not give a larger relative improvement for out-of-domain treebanks. Finally, despite the rich input representations they learn, neural parsers still benefit from structured output prediction of output trees, yielding higher exact match accuracy and stronger generalization both to larger text spans and to out-of-domain corpora. We analyze generalization on English and Chinese corpora, and in the process obtain state-of-the-art parsing results for the Brown, Genia, and English Web treebanks.

In recent NLP research, a topic of interest is universal sentence encoding, sentence representations that can be used in any supervised task. At the word sequence level, fully attention-based models suffer from two problems: a quadratic increase in memory consumption with respect to the sentence length and an inability to capture and use syntactic information. Recursive neural nets can extract very good syntactic information by traversing a tree structure. To this end, we propose Tree Transformer, a model that captures phrase level syntax for constituency trees as well as word-level dependencies for dependency trees by doing recursive traversal only with attention. Evaluation of this model on four tasks gets noteworthy results compared to the standard transformer and LSTM-based models as well as tree-structured LSTMs. Ablation studies to find whether positional information is inherently encoded in the trees and which type of attention is suitable for doing the recursive traversal are provided.

Not all types of supervision signals are created equal: Different types of feedback have different costs and effects on learning. We show how self-regulation strategies that decide when to ask for which kind of feedback from a teacher (or from oneself) can be cast as a learning-to-learn problem leading to improved cost-aware sequence-to-sequence learning. In experiments on interactive neural machine translation, we find that the self-regulator discovers an \epsilon-greedy strategy for the optimal cost-quality trade-off by mixing different feedback types including corrections, error markups, and self-supervision. Furthermore, we demonstrate its robustness under domain shift and identify it as a promising alternative to active learning.

Today, the dominant paradigm for training neural networks involves minimizing task loss on a large dataset. Using world knowledge to inform a model, and yet retain the ability to perform end-to-end training remains an open question. In this paper, we present a novel framework for introducing declarative knowledge to neural network architectures in order to guide training and prediction. Our framework systematically compiles logical statements into computation graphs that augment a neural network without extra learnable parameters or manual redesign. We evaluate our modeling strategy on three tasks: machine comprehension, natural language inference, and text chunking. Our experiments show that knowledge-augmented networks can strongly improve over baselines, especially in low-data regimes.

Pretrained contextual and non-contextual subword embeddings have become available in over 250 languages, allowing massively multilingual NLP. However, while there is no dearth of pretrained embeddings, the distinct lack of systematic evaluations makes it difficult for practitioners to choose between them. In this work, we conduct an extensive evaluation comparing non-contextual subword embeddings, namely FastText and BPEmb, and a contextual representation method, namely BERT, on multilingual named entity recognition and part-of-speech tagging. We find that overall, a combination of BERT, BPEmb, and character representations works best across languages and tasks. A more detailed analysis reveals different strengths and weaknesses: Multilingual BERT performs well in medium- to high-resource languages, but is outperformed by non-contextual subword embeddings in a low-resource setting.

Link prediction is critical for the application of incomplete knowledge graph (KG) in the downstream tasks. As a family of effective approaches for link predictions, embedding methods try to learn low-rank representations for both entities and relations such that the bilinear form defined therein is a well-behaved scoring function. Despite of their successful performances, existing bilinear forms overlook the modeling of relation compositions, resulting in lacks of interpretability for reasoning on KG. To fulfill this gap, we propose a new model called DihEdral, named after dihedral symmetry group. This new model learns knowledge graph embeddings that can capture relation compositions by nature. Furthermore, our approach models the relation embeddings parametrized by discrete values, thereby decrease the solution space drastically. Our experiments show that DihEdral is able to capture all desired properties such as (skew-) symmetry, inversion and (non-) Abelian composition, and outperforms existing bilinear form based approach and is comparable to or better than deep learning models such as ConvE.

Spatial aggregation refers to merging of documents created at the same spatial location. We show that by spatial aggregation of a large collection of documents and applying a traditional topic discovery algorithm on the aggregated data we can efficiently discover spatially distinct topics. By looking at topic discovery through matrix factorization lenses we show that spatial aggregation allows low rank approximation of the original document-word matrix, in which spatially distinct topics are preserved and non-spatial topics are aggregated into a single topic. Our experiments on synthetic data confirm this observation. Our experiments on 4.7 million tweets collected during the Sandy Hurricane in 2012 show that spatial and temporal aggregation allows rapid discovery of relevant spatial and temporal topics during that period. Our work indicates that different forms of document aggregation might be effective in rapid discovery of various types of distinct topics from large collections of documents.

Dependency trees convey rich structural information that is proven useful for extracting relations among entities in text. However, how to effectively make use of relevant information while ignoring irrelevant information from the dependency trees remains a challenging research question. Existing approaches employing rule based hard-pruning strategies for selecting relevant partial dependency structures may not always yield optimal results. In this work, we propose Attention Guided Graph Convolutional Networks (AGGCNs), a novel model which directly takes full dependency trees as inputs. Our model can be understood as a soft-pruning approach that automatically learns how to selectively attend to the relevant sub-structures useful for the relation extraction task. Extensive results on various tasks including cross-sentence n-ary relation extraction and large-scale sentence-level relation extraction show that our model is able to better leverage the structural information of the full dependency trees, giving significantly better results than previous approaches.

We study relation extraction for knowledge base (KB) enrichment. Specifically, we aim to extract entities and their relationships from sentences in the form of triples and map the elements of the extracted triples to an existing KB in an end-to-end manner. Previous studies focus on the extraction itself and rely on Named Entity Disambiguation (NED) to map triples into the KB space. This way, NED errors may cause extraction errors that affect the overall precision and recall. To address this problem, we propose an end-to-end relation extraction model for KB enrichment based on a neural encoder-decoder model. We collect high-quality training data by distant supervision with co-reference resolution and paraphrase detection. We propose an n-gram based attention model that captures multi-word entity names in a sentence. Our model employs jointly learned word and entity embeddings to support named entity disambiguation. Finally, our model uses a modified beam search and a triple classifier to help generate high-quality triples. Our model outperforms state-of-the-art baselines by 15.51% and 8.38% in terms of F1 score on two real-world datasets.

We investigate adaptive ensemble weighting for Neural Machine Translation, addressing the case of improving performance on a new and potentially unknown domain without sacrificing performance on the original domain. We adapt sequentially across two Spanish-English and three English-German tasks, comparing unregularized fine-tuning, L2 and Elastic Weight Consolidation. We then report a novel scheme for adaptive NMT ensemble decoding by extending Bayesian Interpolation with source information, and report strong improvements across test domains without access to the domain label.

It has been shown that the performance of neural machine translation (NMT) drops starkly in low-resource conditions, underperforming phrase-based statistical machine translation (PBSMT) and requiring large amounts of auxiliary data to achieve competitive results. In this paper, we re-assess the validity of these results, arguing that they are the result of lack of system adaptation to low-resource settings. We discuss some pitfalls to be aware of when training low-resource NMT systems, and recent techniques that have shown to be especially helpful in low-resource settings, resulting in a set of best practices for low-resource NMT. In our experiments on German–English with different amounts of IWSLT14 training data, we show that, without the use of any auxiliary monolingual or multilingual data, an optimized NMT system can outperform PBSMT with far less data than previously claimed. We also apply these techniques to a low-resource Korean–English dataset, surpassing previously reported results by 4 BLEU.