COLING 2018 Papers

Complex questions in reading comprehension tasks require integrating information from multiple sentences. In this work, to answer such questions involving temporal and causal relations, we generate event graphs from text based on dependencies, and rank answers by aligning event graphs. In particular, the alignments are constrained by graph-based reasoning to ensure temporal and causal agreement. Our focused approach self-adaptively complements existing solutions; it is automatically triggered only when applicable. Experiments on RACE and MCTest show that state-of-the-art methods are notably improved by using our approach as an add-on.

Neural sequence-to-sequence models have been successfully extended for summary generation. However, existing frameworks generate a single summary for a given input and do not tune the summaries towards any additional constraints/preferences. Such a tunable framework is desirable to account for linguistic preferences of the specific audience who will consume the summary. In this paper, we propose a neural framework to generate summaries constrained to a vocabulary-defined linguistic preferences of a target audience. The proposed method accounts for the generation context by tuning the summary words at the time of generation. Our evaluations indicate that the proposed approach tunes summaries to the target vocabulary while still maintaining a superior summary quality against a state-of-the-art word embedding based lexical substitution algorithm, suggesting the feasibility of the proposed approach. We demonstrate two applications of the proposed approach - to generate understandable summaries with simpler words, and readable summaries with shorter words.

One main challenge for incremental transition-based parsers, when future inputs are invisible, is to extract good features from a limited local context. In this work, we present a simple technique to maximally utilize the local features with an attention mechanism, which works as context- dependent dynamic feature selection. Our model learns, for example, which tokens should a parser focus on, to decide the next action. Our multilingual experiment shows its effectiveness across many languages. We also present an experiment with augmented test dataset and demon- strate it helps to understand the model’s behavior on locally ambiguous points.

Aspect-level sentiment analysis aims to distinguish the sentiment polarity of each specific aspect term in a given sentence. Both industry and academia have realized the importance of the relationship between aspect term and sentence, and made attempts to model the relationship by designing a series of attention models. However, most existing methods usually neglect the fact that the position information is also crucial for identifying the sentiment polarity of the aspect term. When an aspect term occurs in a sentence, its neighboring words should be given more attention than other words with long distance. Therefore, we propose a position-aware bidirectional attention network (PBAN) based on bidirectional GRU. PBAN not only concentrates on the position information of aspect terms, but also mutually models the relation between aspect term and sentence by employing bidirectional attention mechanism. The experimental results on SemEval 2014 Datasets demonstrate the effectiveness of our proposed PBAN model.

Self-normalizing discriminative models approximate the normalized probability of a class without having to compute the partition function. In the context of language modeling, this property is particularly appealing as it may significantly reduce run-times due to large word vocabularies. In this study, we provide a comprehensive investigation of language modeling self-normalization. First, we theoretically analyze the inherent self-normalization properties of Noise Contrastive Estimation (NCE) language models. Then, we compare them empirically to softmax-based approaches, which are self-normalized using explicit regularization, and suggest a hybrid model with compelling properties. Finally, we uncover a surprising negative correlation between self-normalization and perplexity across the board, as well as some regularity in the observed errors, which may potentially be used for improving self-normalization algorithms in the future.

This paper presents RuSentiment, a new dataset for sentiment analysis of social media posts in Russian, and a new set of comprehensive annotation guidelines that are extensible to other languages. RuSentiment is currently the largest in its class for Russian, with 31,185 posts annotated with Fleiss’ kappa of 0.58 (3 annotations per post). To diversify the dataset, 6,950 posts were pre-selected with an active learning-style strategy. We report baseline classification results, and we also release the best-performing embeddings trained on 3.2B tokens of Russian VKontakte posts.

Because of license restrictions, it often becomes impossible to strictly reproduce most research results on Twitter data already a few months after the creation of the corpus. This situation worsened gradually as time passes and tweets become inaccessible. This is a critical issue for reproducible and accountable research on social media. We partly solve this challenge by annotating a new Twitter-like corpus from an alternative large social medium with licenses that are compatible with reproducible experiments: Mastodon. We manually annotate both dialogues and sentiments on this corpus, and train a multi-task hierarchical recurrent network on joint sentiment and dialog act recognition. We experimentally demonstrate that transfer learning may be efficiently achieved between both tasks, and further analyze some specific correlations between sentiments and dialogues on social media. Both the annotated corpus and deep network are released with an open-source license.

This paper proposes an extractive multi-document summarization approach based on an ant colony system to optimize the information coverage of summary sentences. The implemented system was evaluated on both English and Arabic versions of the corpus of the Text Analysis Conference 2011 MultiLing Pilot by using ROUGE metrics. The evaluation results are promising in comparison to those of the participating systems. Indeed, our system achieved the best scores based on several ROUGE metrics.

Existing neural models usually predict the tag of the current token independent of the neighboring tags. The popular LSTM-CRF model considers the tag dependencies between every two consecutive tags. However, it is hard for existing neural models to take longer distance dependencies between tags into consideration. The scalability is mainly limited by the complex model structures and the cost of dynamic programming during training. In our work, we first design a new model called “high order LSTM” to predict multiple tags for the current token which contains not only the current tag but also the previous several tags. We call the number of tags in one prediction as “order”. Then we propose a new method called Multi-Order BiLSTM (MO-BiLSTM) which combines low order and high order LSTMs together. MO-BiLSTM keeps the scalability to high order models with a pruning technique. We evaluate MO-BiLSTM on all-phrase chunking and NER datasets. Experiment results show that MO-BiLSTM achieves the state-of-the-art result in chunking and highly competitive results in two NER datasets.

Named entity recognition (NER) has attracted a substantial amount of research. Recently, several neural network-based models have been proposed and achieved high performance. However, there is little research on fine-grained NER (FG-NER), in which hundreds of named entity categories must be recognized, especially for non-English languages. It is still an open question whether there is a model that is robust across various settings or the proper model varies depending on the language, the number of named entity categories, and the size of training datasets. This paper first presents an empirical comparison of FG-NER models for English and Japanese and demonstrates that LSTM+CNN+CRF (Ma and Hovy, 2016), one of the state-of-the-art methods for English NER, also works well for English FG-NER but does not work well for Japanese, a language that has a large number of character types. To tackle this problem, we propose a method to improve the neural network-based Japanese FG-NER performance by removing the CNN layer and utilizing dictionary and category embeddings. Experiment results show that the proposed method improves Japanese FG-NER F-score from 66.76% to 75.18%.

This work addresses challenges arising from extracting entities from textual data, including the high cost of data annotation, model accuracy, selecting appropriate evaluation criteria, and the overall quality of annotation. We present a framework that integrates Entity Set Expansion (ESE) and Active Learning (AL) to reduce the annotation cost of sparse data and provide an online evaluation method as feedback. This incremental and interactive learning framework allows for rapid annotation and subsequent extraction of sparse data while maintaining high accuracy. We evaluate our framework on three publicly available datasets and show that it drastically reduces the cost of sparse entity annotation by an average of 85% and 45% to reach 0.9 and 1.0 F-Scores respectively. Moreover, the method exhibited robust performance across all datasets.

Fundamental to several knowledge-centric applications is the need to identify named entities from their textual mentions. However, entities lack a unique representation and their mentions can differ greatly. These variations arise in complex ways that cannot be captured using textual similarity metrics. However, entities have underlying structures, typically shared by entities of the same entity type, that can help reason over their name variations. Discovering, learning and manipulating these structures typically requires high manual effort in the form of large amounts of labeled training data and handwritten transformation programs. In this work, we propose an active-learning based framework that drastically reduces the labeled data required to learn the structures of entities. We show that programs for mapping entity mentions to their structures can be automatically generated using human-comprehensible labels. Our experiments show that our framework consistently outperforms both handwritten programs and supervised learning models. We also demonstrate the utility of our framework in relation extraction and entity resolution tasks.

Entity Linking aims to link entity mentions in texts to knowledge bases, and neural models have achieved recent success in this task. However, most existing methods rely on local contexts to resolve entities independently, which may usually fail due to the data sparsity of local information. To address this issue, we propose a novel neural model for collective entity linking, named as NCEL. NCEL apply Graph Convolutional Network to integrate both local contextual features and global coherence information for entity linking. To improve the computation efficiency, we approximately perform graph convolution on a subgraph of adjacent entity mentions instead of those in the entire text. We further introduce an attention scheme to improve the robustness of NCEL to data noise and train the model on Wikipedia hyperlinks to avoid overfitting and domain bias. In experiments, we evaluate NCEL on five publicly available datasets to verify the linking performance as well as generalization ability. We also conduct an extensive analysis of time complexity, the impact of key modules, and qualitative results, which demonstrate the effectiveness and efficiency of our proposed method.

State-of-the-art entity linkers achieve high accuracy scores with probabilistic methods. However, these scores should be considered in relation to the properties of the datasets they are evaluated on. Until now, there has not been a systematic investigation of the properties of entity linking datasets and their impact on system performance. In this paper we report on a series of hypotheses regarding the long tail phenomena in entity linking datasets, their interaction, and their impact on system performance. Our systematic study of these hypotheses shows that evaluation datasets mainly capture head entities and only incidentally cover data from the tail, thus encouraging systems to overfit to popular/frequent and non-ambiguous cases. We find the most difficult cases of entity linking among the infrequent candidates of ambiguous forms. With our findings, we hope to inspire future designs of both entity linking systems and evaluation datasets. To support this goal, we provide a list of recommended actions for better inclusion of tail cases.

Evaluating on adversarial examples has become a standard procedure to measure robustness of deep learning models. Due to the difficulty of creating white-box adversarial examples for discrete text input, most analyses of the robustness of NLP models have been done through black-box adversarial examples. We investigate adversarial examples for character-level neural machine translation (NMT), and contrast black-box adversaries with a novel white-box adversary, which employs differentiable string-edit operations to rank adversarial changes. We propose two novel types of attacks which aim to remove or change a word in a translation, rather than simply break the NMT. We demonstrate that white-box adversarial examples are significantly stronger than their black-box counterparts in different attack scenarios, which show more serious vulnerabilities than previously known. In addition, after performing adversarial training, which takes only 3 times longer than regular training, we can improve the model’s robustness significantly.

Recently, neural machine translation (NMT) has been extended to multilinguality, that is to handle more than one translation direction with a single system. Multilingual NMT showed competitive performance against pure bilingual systems. Notably, in low-resource settings, it proved to work effectively and efficiently, thanks to shared representation space that is forced across languages and induces a sort of transfer-learning. Furthermore, multilingual NMT enables so-called zero-shot inference across language pairs never seen at training time. Despite the increasing interest in this framework, an in-depth analysis of what a multilingual NMT model is capable of and what it is not is still missing. Motivated by this, our work (i) provides a quantitative and comparative analysis of the translations produced by bilingual, multilingual and zero-shot systems; (ii) investigates the translation quality of two of the currently dominant neural architectures in MT, which are the Recurrent and the Transformer ones; and (iii) quantitatively explores how the closeness between languages influences the zero-shot translation. Our analysis leverages multiple professional post-edits of automatic translations by several different systems and focuses both on automatic standard metrics (BLEU and TER) and on widely used error categories, which are lexical, morphology, and word order errors.

We analyze some of the fundamental design challenges that impact the development of a multilingual state-of-the-art named entity transliteration system, including curating bilingual named entity datasets and evaluation of multiple transliteration methods. We empirically evaluate the transliteration task using the traditional weighted finite state transducer (WFST) approach against two neural approaches: the encoder-decoder recurrent neural network method and the recent, non-sequential Transformer method. In order to improve availability of bilingual named entity transliteration datasets, we release personal name bilingual dictionaries mined from Wikidata for English to Russian, Hebrew, Arabic, and Japanese Katakana. Our code and dictionaries are publicly available.

This paper focuses on subword-based Neural Machine Translation (NMT). We hypothesize that in the NMT model, the appropriate subword units for the following three modules (layers) can differ: (1) the encoder embedding layer, (2) the decoder embedding layer, and (3) the decoder output layer. We find the subword based on Sennrich et al. (2016) has a feature that a large vocabulary is a superset of a small vocabulary and modify the NMT model enables the incorporation of several different subword units in a single embedding layer. We refer these small subword features as hierarchical subword features. To empirically investigate our assumption, we compare the performance of several different subword units and hierarchical subword features for both the encoder and decoder embedding layers. We confirmed that incorporating hierarchical subword features in the encoder consistently improves BLEU scores on the IWSLT evaluation datasets.

Neural machine translation (NMT) systems are usually trained on a large amount of bilingual sentence pairs and translate one sentence at a time, ignoring inter-sentence information. This may make the translation of a sentence ambiguous or even inconsistent with the translations of neighboring sentences. In order to handle this issue, we propose an inter-sentence gate model that uses the same encoder to encode two adjacent sentences and controls the amount of information flowing from the preceding sentence to the translation of the current sentence with an inter-sentence gate. In this way, our proposed model can capture the connection between sentences and fuse recency from neighboring sentences into neural machine translation. On several NIST Chinese-English translation tasks, our experiments demonstrate that the proposed inter-sentence gate model achieves substantial improvements over the baseline.