COLING 2016 论文列表

Knowledge base (KB) such as Freebase plays an important role for many natural language processing tasks. English knowledge base is obviously larger and of higher quality than low resource language like Chinese. To expand Chinese KB by leveraging English KB resources, an effective way is to translate English KB (source) into Chinese (target). In this direction, two major challenges are to model triple semantics and to build a robust KB translator. We address these challenges by presenting a neural network approach, which learns continuous triple representation with a gated neural network. Accordingly, source triples and target triples are mapped in the same semantic vector space. We build a new dataset for English-Chinese KB translation from Freebase, and compare with several baselines on it. Experimental results show that the proposed method improves translation accuracy compared with baseline methods. We show that adaptive composition model improves standard solution such as neural tensor network in terms of translation accuracy.

In this paper, we propose a novel neural approach for paraphrase generation. Conventional paraphrase generation methods either leverage hand-written rules and thesauri-based alignments, or use statistical machine learning principles. To the best of our knowledge, this work is the first to explore deep learning models for paraphrase generation. Our primary contribution is a stacked residual LSTM network, where we add residual connections between LSTM layers. This allows for efficient training of deep LSTMs. We evaluate our model and other state-of-the-art deep learning models on three different datasets: PPDB, WikiAnswers, and MSCOCO. Evaluation results demonstrate that our model outperforms sequence to sequence, attention-based, and bi-directional LSTM models on BLEU, METEOR, TER, and an embedding-based sentence similarity metric.

We describe an attentive encoder that combines tree-structured recursive neural networks and sequential recurrent neural networks for modelling sentence pairs. Since existing attentive models exert attention on the sequential structure, we propose a way to incorporate attention into the tree topology. Specially, given a pair of sentences, our attentive encoder uses the representation of one sentence, which generated via an RNN, to guide the structural encoding of the other sentence on the dependency parse tree. We evaluate the proposed attentive encoder on three tasks: semantic similarity, paraphrase identification and true-false question selection. Experimental results show that our encoder outperforms all baselines and achieves state-of-the-art results on two tasks.

Inference rule discovery aims to identify entailment relations between predicates, e.g., ‘X acquire Y –> X purchase Y’ and ‘X is author of Y –> X write Y’. Traditional methods dis-cover inference rules by computing distributional similarities between predicates, with each predicate is represented as one or more feature vectors of its instantiations. These methods, however, have two main drawbacks. Firstly, these methods are mostly context-insensitive, cannot accurately measure the similarity between two predicates in a specific context. Secondly, traditional methods usually model predicates independently, ignore the rich inter-dependencies between predicates. To address the above two issues, this pa-per proposes a graph-based method, which can discover inference rules by effectively modelling and exploiting both the context and the inter-dependencies between predicates. Specifically, we propose a graph-based representation—Predicate Graph, which can capture the semantic relevance between predicates using both the predicate-feature co-occurrence statistics and the inter-dependencies between predicates. Based on the predicate graph, we propose a context-sensitive random walk algorithm, which can learn con-text-specific predicate representations by distinguishing context-relevant information from context-irrelevant information. Experimental results show that our method significantly outperforms traditional inference rule discovery methods.

Sentence intersection captures the semantic overlap of two texts, generalizing over paradigms such as textual entailment and semantic text similarity. Despite its modeling power, it has received little attention because it is difficult for non-experts to annotate. We analyze 200 pairs of similar sentences and identify several underlying properties of sentence intersection. We leverage these insights to design an algorithm that decomposes the sentence intersection task into several simpler annotation tasks, facilitating the construction of a high quality dataset via crowdsourcing. We implement this approach and provide an annotated dataset of 1,764 sentence intersections.

Existing systems deliver high accuracy and F1-scores for detecting paraphrase and semantic similarity on traditional clean-text corpus. For instance, on the clean-text Microsoft Paraphrase benchmark database, the existing systems attain an accuracy as high as 0:8596. However, existing systems for detecting paraphrases and semantic similarity on user-generated short-text content on microblogs such as Twitter, comprising of noisy and ad hoc short-text, needs significant research attention. In this paper, we propose a machine learning based approach towards this. We propose a set of features that, although well-known in the NLP literature for solving other problems, have not been explored for detecting paraphrase or semantic similarity, on noisy user-generated short-text data such as Twitter. We apply support vector machine (SVM) based learning. We use the benchmark Twitter paraphrase data, released as a part of SemEval 2015, for experiments. Our system delivers a paraphrase detection F1-score of 0.717 and semantic similarity detection F1-score of 0.741, thereby significantly outperforming the existing systems, that deliver F1-scores of 0.696 and 0.724 for the two problems respectively. Our features also allow us to obtain a rank among the top-10, when trained on the Microsoft Paraphrase corpus and tested on the corresponding test data, thereby empirically establishing our approach as ubiquitous across the different paraphrase detection databases.

Recognizing Textual Entailment (RTE) is a fundamentally important task in natural language processing that has many applications. The recently released Stanford Natural Language Inference (SNLI) corpus has made it possible to develop and evaluate deep neural network methods for the RTE task. Previous neural network based methods usually try to encode the two sentences (premise and hypothesis) and send them together into a multi-layer perceptron to get their entailment type, or use LSTM-RNN to link two sentences together while using attention mechanic to enhance the model’s ability. In this paper, we propose to use the re-read mechanic, which means to read the premise again and again while reading the hypothesis. After read the premise again, the model can get a better understanding of the premise, which can also affect the understanding of the hypothesis. On the contrary, a better understanding of the hypothesis can also affect the understanding of the premise. With the alternative re-read process, the model can “think” of a better decision of entailment type. We designed a new LSTM unit called re-read LSTM (rLSTM) to implement this “thinking” process. Experiments show that we achieve results better than current state-of-the-art equivalents.

In this paper, we propose utilising eye gaze information for estimating parameters of a Japanese predicate argument structure (PAS) analysis model. We employ not only linguistic information in the text, but also the information of annotator eye gaze during their annotation process. We hypothesise that annotator’s frequent looks at certain candidates imply their plausibility of being the argument of the predicate. Based on this hypothesis, we consider annotator eye gaze for estimating the model parameters of the PAS analysis. The evaluation experiment showed that introducing eye gaze information increased the accuracy of the PAS analysis by 0.05 compared with the conventional methods.

According to the distributional inclusion hypothesis, entailment between words can be measured via the feature inclusions of their distributional vectors. In recent work, we showed how this hypothesis can be extended from words to phrases and sentences in the setting of compositional distributional semantics. This paper focuses on inclusion properties of tensors; its main contribution is a theoretical and experimental analysis of how feature inclusion works in different concrete models of verb tensors. We present results for relational, Frobenius, projective, and holistic methods and compare them to the simple vector addition, multiplication, min, and max models. The degrees of entailment thus obtained are evaluated via a variety of existing word-based measures, such as Weed’s and Clarke’s, KL-divergence, APinc, balAPinc, and two of our previously proposed metrics at the phrase/sentence level. We perform experiments on three entailment datasets, investigating which version of tensor-based composition achieves the highest performance when combined with the sentence-level measures.

The paper presents an iterative bidirectional clustering of adjectives and nouns based on a co-occurrence matrix. The clustering method combines a Vector Space Models (VSM) and the results of a Latent Dirichlet Allocation (LDA), whose results are merged in each iterative step. The aim is to derive a clustering of German adjectives that reflects latent semantic classes of adjectives, and that can be used to induce frame-based representations of nouns in a later step. We are able to show that the method induces meaningful groups of adjectives, and that it outperforms a baseline k-means algorithm.

This paper proposes a novel problem setting of selectional preference (SP) between a predicate and its arguments, called as context-sensitive SP (CSP). CSP models the narrative consistency between the predicate and preceding contexts of its arguments, in addition to the conventional SP based on semantic types. Furthermore, we present a novel CSP model that extends the neural SP model (Van de Cruys, 2014) to incorporate contextual information into the distributed representations of arguments. Experimental results demonstrate that the proposed CSP model successfully learns CSP and outperforms the conventional SP model in coreference cluster ranking.

Temporal relation classification is a challenging task, especially when there are no explicit markers to characterise the relation between temporal entities. This occurs frequently in inter-sentential relations, whose entities are not connected via direct syntactic relations making classification even more difficult. In these cases, resorting to features that focus on the semantic content of the event words may be very beneficial for inferring implicit relations. Specifically, while morpho-syntactic and context features are considered sufficient for classifying event-timex pairs, we believe that exploiting distributional semantic information about event words can benefit supervised classification of other types of pairs. In this work, we assess the impact of using word embeddings as features for event words in classifying temporal relations of event-event pairs and event-DCT (document creation time) pairs.

Human concept representations are often grounded with visual information, yet some aspects of meaning cannot be visually represented or are better described with language. Thus, vision and language provide complementary information that, properly combined, can potentially yield more complete concept representations. Recently, state-of-the-art distributional semantic models and convolutional neural networks have achieved great success in representing linguistic and visual knowledge respectively. In this paper, we compare both, visual and linguistic representations in their ability to capture different types of fine-grain semantic knowledge—or attributes—of concepts. Humans often describe objects using attributes, that is, properties such as shape, color or functionality, which often transcend the linguistic and visual modalities. In our setting, we evaluate how well attributes can be predicted by using the unimodal representations as inputs. We are interested in first, finding out whether attributes are generally better captured by either the vision or by the language modality; and second, if none of them is clearly superior (as we hypothesize), what type of attributes or semantic knowledge are better encoded from each modality. Ultimately, our study sheds light on the potential of combining visual and textual representations.

In this paper, we outline an approach to build graph-based reverse dictionaries using word definitions. A reverse dictionary takes a phrase as an input and outputs a list of words semantically similar to that phrase. It is a solution to the Tip-of-the-Tongue problem. We use a distance-based similarity measure, computed on a graph, to assess the similarity between a word and the input phrase. We compare the performance of our approach with the Onelook Reverse Dictionary and a distributional semantics method based on word2vec, and show that our approach is much better than the distributional semantics method, and as good as Onelook, on a 3k lexicon. This simple approach sets a new performance baseline for reverse dictionaries.

We assess the reliability and accuracy of (neural) word embeddings for both modern and historical English and German. Our research provides deeper insights into the empirically justified choice of optimal training methods and parameters. The overall low reliability we observe, nevertheless, casts doubt on the suitability of word neighborhoods in embedding spaces as a basis for qualitative conclusions on synchronic and diachronic lexico-semantic matters, an issue currently high up in the agenda of Digital Humanities.

The recent proliferation of smart devices necessitates methods to learn small-sized models. This paper demonstrates that if there are m features in total but only n = o(\sqrt{m}) features are required to distinguish examples, with \Omega(\log m) training examples and reasonable settings, it is possible to obtain a good model in a succinct representation using n \log_2 \frac{m}{n} + o(m) bits, by using a pipeline of existing compression methods: L1-regularized logistic regression, feature hashing, Elias–Fano indices, and randomized quantization. An experiment shows that a noun phrase chunking task for which an existing library requires 27 megabytes can be compressed to less than 13 kilobytes without notable loss of accuracy.

Several tasks in argumentation mining and debating, question-answering, and natural language inference involve classifying a sequence in the context of another sequence (referred as bi-sequence classification). For several single sequence classification tasks, the current state-of-the-art approaches are based on recurrent and convolutional neural networks. On the other hand, for bi-sequence classification problems, there is not much understanding as to the best deep learning architecture. In this paper, we attempt to get an understanding of this category of problems by extensive empirical evaluation of 19 different deep learning architectures (specifically on different ways of handling context) for various problems originating in natural language processing like debating, textual entailment and question-answering. Following the empirical evaluation, we offer our insights and conclusions regarding the architectures we have considered. We also establish the first deep learning baselines for three argumentation mining tasks.

Some expressions can be ambiguous between idiomatic and literal interpretations depending on the context they occur in, e.g., ‘sales hit the roof’ vs. ‘hit the roof of the car’. We present a novel method of classifying whether a given instance is literal or idiomatic, focusing on verb-noun constructions. We report state-of-the-art results on this task using an approach based on the hypothesis that the distributions of the contexts of the idiomatic phrases will be different from the contexts of the literal usages. We measure contexts by using projections of the words into vector space. For comparison, we implement Fazly et al. (2009)’s, Sporleder and Li (2009)’s, and Li and Sporleder (2010b)’s methods and apply them to our data. We provide experimental results validating the proposed techniques.

When processing arguments in online user interactive discourse, it is often necessary to determine their bases of support. In this paper, we describe a supervised approach, based on deep neural networks, for classifying the claims made in online arguments. We conduct experiments using convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) on two claim data sets compiled from online user comments. Using different types of distributional word embeddings, but without incorporating any rich, expensive set of features, we achieve a significant improvement over the state of the art for one data set (which categorizes arguments as factual vs. emotional), and performance comparable to the state of the art on the other data set (which categorizes propositions according to their verifiability). Our approach has the advantages of using a generalized, simple, and effective methodology that works for claim categorization on different data sets and tasks.