COLING 2016 Papers

We propose a novel word embedding-based hypernym generation model that jointly learns clusters of hyponym-hypernym relations, i.e., hypernymy, and projections from hyponym to hypernym embeddings. Most of the recent hypernym detection models focus on a hypernymy classification problem that determines whether a pair of words is in hypernymy or not. These models do not directly deal with a hypernym generation problem in that a model generates hypernyms for a given word. Differently from previous studies, our model jointly learns the clusters and projections with adjusting the number of clusters so that the number of clusters can be determined depending on the learned projections and vice versa. Our model also boosts the performance by incorporating inner product-based similarity measures and negative examples, i.e., sampled non-hypernyms, into our objectives in learning. We evaluated our joint learning models on the task of Japanese and English hypernym generation and showed a significant improvement over an existing pipeline model. Our model also compared favorably to existing distributed hypernym detection models on the English hypernym classification task.

Most distributional lexico-semantic models derive their representations based on external language resources such as text corpora. In this study, we propose that internal language models, that are more closely aligned to the mental representations of words could provide important insights into cognitive science, including linguistics. Doing so allows us to reflect upon theoretical questions regarding the structure of the mental lexicon, and also puts into perspective a number of assumptions underlying recently proposed distributional text-based models. In particular, we focus on word-embedding models which have been proposed to learn aspects of word meaning in a manner similar to humans. These are contrasted with internal language models derived from a new extensive data set of word associations. Using relatedness and similarity judgments we evaluate these models and find that the word-association-based internal language models consistently outperform current state-of-the art text-based external language models, often with a large margin. These results are not just a performance improvement; they also have implications for our understanding of how distributional knowledge is used by people.

We propose a new word embedding model, inspired by GloVe, which is formulated as a feasible least squares optimization problem. In contrast to existing models, we explicitly represent the uncertainty about the exact definition of each word vector. To this end, we estimate the error that results from using noisy co-occurrence counts in the formulation of the model, and we model the imprecision that results from including uninformative context words. Our experimental results demonstrate that this model compares favourably with existing word embedding models.

With the success of word embedding methods in various Natural Language Processing tasks, all the field of distributional semantics has experienced a renewed interest. Beside the famous word2vec, recent studies have presented efficient techniques to build distributional thesaurus; in particular, Claveau et al. (2014) have already shown that Information Retrieval (IR) tools and concepts can be successfully used to build a thesaurus. In this paper, we address the problem of the evaluation of such thesauri or embedding models and compare their results. Through several experiments and by evaluating directly the results with reference lexicons, we show that the recent IR-based distributional models outperform state-of-the-art systems such as word2vec. Following the work of Claveau and Kijak (2016), we use IR as an applicative framework to indirectly evaluate the generated thesaurus. Here again, this task-based evaluation validates the IR approach used to build the thesaurus. Moreover, it allows us to compare these results with those from the direct evaluation framework used in the literature. The observed differences bring these evaluation habits into question.

Recently, the development of neural machine translation (NMT) has significantly improved the translation quality of automatic machine translation. While most sentences are more accurate and fluent than translations by statistical machine translation (SMT)-based systems, in some cases, the NMT system produces translations that have a completely different meaning. This is especially the case when rare words occur. When using statistical machine translation, it has already been shown that significant gains can be achieved by simplifying the input in a preprocessing step. A commonly used example is the pre-reordering approach. In this work, we used phrase-based machine translation to pre-translate the input into the target language. Then a neural machine translation system generates the final hypothesis using the pre-translation. Thereby, we use either only the output of the phrase-based machine translation (PBMT) system or a combination of the PBMT output and the source sentence. We evaluate the technique on the English to German translation task. Using this approach we are able to outperform the PBMT system as well as the baseline neural MT system by up to 2 BLEU points. We analyzed the influence of the quality of the initial system on the final result.

We introduce a distribution based model to learn bilingual word embeddings from monolingual data. It is simple, effective and does not require any parallel data or any seed lexicon. We take advantage of the fact that word embeddings are usually in form of dense real-valued low-dimensional vector and therefore the distribution of them can be accurately estimated. A novel cross-lingual learning objective is proposed which directly matches the distributions of word embeddings in one language with that in the other language. During the joint learning process, we dynamically estimate the distributions of word embeddings in two languages respectively and minimize the dissimilarity between them through standard back propagation algorithm. Our learned bilingual word embeddings allow to group each word and its translations together in the shared vector space. We demonstrate the utility of the learned embeddings on the task of finding word-to-word translations from monolingual corpora. Our model achieved encouraging performance on data in both related languages and substantially different languages.

In recent years, neural machine translation (NMT) has demonstrated state-of-the-art machine translation (MT) performance. It is a new approach to MT, which tries to learn a set of parameters to maximize the conditional probability of target sentences given source sentences. In this paper, we present a novel approach to improve the translation performance in NMT by conveying topic knowledge during translation. The proposed topic-informed NMT can increase the likelihood of selecting words from the same topic and domain for translation. Experimentally, we demonstrate that topic-informed NMT can achieve a 1.15 (3.3% relative) and 1.67 (5.4% relative) absolute improvement in BLEU score on the Chinese-to-English language pair using NIST 2004 and 2005 test sets, respectively, compared to NMT without topic information.

Interactive-predictive machine translation (IPMT) is a translation mode which combines machine translation technology and human behaviours. In the IPMT system, the utilization of the prefix greatly affects the interaction efficiency. However, state-of-the-art methods filter translation hypotheses mainly according to their matching results with the prefix on character level, and the advantage of the prefix is not fully developed. Focusing on this problem, this paper mines the deep constraints of prefix on syntactic level to improve the performance of IPMT systems. Two syntactic subtree matching rules based on phrase structure grammar are proposed to filter the translation hypotheses more strictly. Experimental results on LDC Chinese-English corpora show that the proposed method outperforms state-of-the-art phrase-based IPMT system while keeping comparable decoding speed.

In modern text annotation projects, crowdsourced annotations are often aggregated using item response models or by majority vote. Recently, item response models enhanced with generative data models have been shown to yield substantial benefits over those with conditional or no data models. However, suitable generative data models do not exist for many tasks, such as semantic labeling tasks. When no generative data model exists, we demonstrate that similar benefits may be derived by conditionally modeling documents that have been previously embedded in a semantic space using recent work in vector space models. We use this approach to show state-of-the-art results on a variety of semantic annotation aggregation tasks.

Reading comprehension has embraced a booming in recent NLP research. Several institutes have released the Cloze-style reading comprehension data, and these have greatly accelerated the research of machine comprehension. In this work, we firstly present Chinese reading comprehension datasets, which consist of People Daily news dataset and Children’s Fairy Tale (CFT) dataset. Also, we propose a consensus attention-based neural network architecture to tackle the Cloze-style reading comprehension problem, which aims to induce a consensus attention over every words in the query. Experimental results show that the proposed neural network significantly outperforms the state-of-the-art baselines in several public datasets. Furthermore, we setup a baseline for Chinese reading comprehension task, and hopefully this would speed up the process for future research.

The exchangeability assumption in topic models like Latent Dirichlet Allocation (LDA) often results in inferring inconsistent topics for the words of text spans like noun-phrases, which are usually expected to be topically coherent. We propose copulaLDA, that extends LDA by integrating part of the text structure to the model and relaxes the conditional independence assumption between the word-specific latent topics given the per-document topic distributions. To this end, we assume that the words of text spans like noun-phrases are topically bound and we model this dependence with copulas. We demonstrate empirically the effectiveness of copulaLDA on both intrinsic and extrinsic evaluation tasks on several publicly available corpora.

This paper presents a novel approach to recurrent neural network (RNN) regularization. Differently from the widely adopted dropout method, which is applied to forward connections of feedforward architectures or RNNs, we propose to drop neurons directly in recurrent connections in a way that does not cause loss of long-term memory. Our approach is as easy to implement and apply as the regular feed-forward dropout and we demonstrate its effectiveness for the most effective modern recurrent network – Long Short-Term Memory network. Our experiments on three NLP benchmarks show consistent improvements even when combined with conventional feed-forward dropout.

This work focuses on answering single-relation factoid questions over Freebase. Each question can acquire the answer from a single fact of form (subject, predicate, object) in Freebase. This task, simple question answering (SimpleQA), can be addressed via a two-step pipeline: entity linking and fact selection. In fact selection, we match the subject entity in a fact candidate with the entity mention in the question by a character-level convolutional neural network (char-CNN), and match the predicate in that fact with the question by a word-level CNN (word-CNN). This work makes two main contributions. (i) A simple and effective entity linker over Freebase is proposed. Our entity linker outperforms the state-of-the-art entity linker over SimpleQA task. (ii) A novel attentive maxpooling is stacked over word-CNN, so that the predicate representation can be matched with the predicate-focused question representation more effectively. Experiments show that our system sets new state-of-the-art in this task.

In real-world data, e.g., from Web forums, text is often contaminated with redundant or irrelevant content, which leads to introducing noise in machine learning algorithms. In this paper, we apply Long Short-Term Memory networks with an attention mechanism, which can select important parts of text for the task of similar question retrieval from community Question Answering (cQA) forums. In particular, we use the attention weights for both selecting entire sentences and their subparts, i.e., word/chunk, from shallow syntactic trees. More interestingly, we apply tree kernels to the filtered text representations, thus exploiting the implicit features of the subtree space for learning question reranking. Our results show that the attention-based pruning allows for achieving the top position in the cQA challenge of SemEval 2016, with a relatively large gap from the other participants while greatly decreasing running time.

Ordinal regression which is known with learning to rank has long been used in information retrieval (IR). Learning to rank algorithms, have been tailored in document ranking, information filtering, and building large aligned corpora successfully. In this paper, we propose to use this algorithm for query modeling in cross-language environments. To this end, first we build a query-generated training data using pseudo-relevant documents to the query and all translation candidates. The pseudo-relevant documents are obtained by top-ranked documents in response to a translation of the original query. The class of each candidate in the training data is determined based on presence/absence of the candidate in the pseudo-relevant documents. We learn an ordinal regression model to score the candidates based on their relevance to the context of the query, and after that, we construct a query-dependent translation model using a softmax function. Finally, we re-weight the query based on the obtained model. Experimental results on French, German, Spanish, and Italian CLEF collections demonstrate that the proposed method achieves better results compared to state-of-the-art cross-language information retrieval methods, particularly in long queries with large training data.

Knowledge graph (KG) completion adds new facts to a KG by making inferences from existing facts. Most existing methods ignore the time information and only learn from time-unknown fact triples. In dynamic environments that evolve over time, it is important and challenging for knowledge graph completion models to take into account the temporal aspects of facts. In this paper, we present a novel time-aware knowledge graph completion model that is able to predict links in a KG using both the existing facts and the temporal information of the facts. To incorporate the happening time of facts, we propose a time-aware KG embedding model using temporal order information among facts. To incorporate the valid time of facts, we propose a joint time-aware inference model based on Integer Linear Programming (ILP) using temporal consistencyinformationasconstraints. Wefurtherintegratetwomodelstomakefulluseofglobal temporal information. We empirically evaluate our models on time-aware KG completion task. Experimental results show that our time-aware models achieve the state-of-the-art on temporal facts consistently.

We analyze the performance of encoder-decoder neural models and compare them with well-known established methods. The latter represent different classes of traditional approaches that are applied to the monotone sequence-to-sequence tasks OCR post-correction, spelling correction, grapheme-to-phoneme conversion, and lemmatization. Such tasks are of practical relevance for various higher-level research fields including digital humanities, automatic text correction, and speech recognition. We investigate how well generic deep-learning approaches adapt to these tasks, and how they perform in comparison with established and more specialized methods, including our own adaptation of pruned CRFs.

Language students are most engaged while reading texts at an appropriate difficulty level. However, existing methods of evaluating text difficulty focus mainly on vocabulary and do not prioritize grammatical features, hence they do not work well for language learners with limited knowledge of grammar. In this paper, we introduce grammatical templates, the expert-identified units of grammar that students learn from class, as an important feature of text difficulty evaluation. Experimental classification results show that grammatical template features significantly improve text difficulty prediction accuracy over baseline readability features by 7.4%. Moreover,we build a simple and human-understandable text difficulty evaluation approach with 87.7% accuracy, using only 5 grammatical template features.

Argument mining aims to determine the argumentative structure of texts. Although it is said to be crucial for future applications such as writing support systems, the benefit of its output has rarely been evaluated. This paper puts the analysis of the output into the focus. In particular, we investigate to what extent the mined structure can be leveraged to assess the argumentation quality of persuasive essays. We find insightful statistical patterns in the structure of essays. From these, we derive novel features that we evaluate in four argumentation-related essay scoring tasks. Our results reveal the benefit of argument mining for assessing argumentation quality. Among others, we improve the state of the art in scoring an essay’s organization and its argument strength.