ACL 2015 Papers

Traditional approaches to the task of ACE event extraction primarily rely on elaborately designed features and complicated natural language processing (NLP) tools. These traditional approaches lack generalization, take a large amount of human effort and are prone to error propagation and data sparsity problems. This paper proposes a novel event-extraction method, which aims to automatically extract lexical-level and sentence-level features without using complicated NLP tools. We introduce a word-representation model to capture meaningful semantic regularities for words and adopt a framework based on a convolutional neural network (CNN) to capture sentence-level clues. However, CNN can only capture the most important information in a sentence and may miss valuable facts when considering multiple-event sentences. We propose a dynamic multi-pooling convolutional neural network (DMCNN), which uses a dynamic multi-pooling layer according to event triggers and arguments, to reserve more crucial information. The experimental results show that our approach significantly outperforms other state-of-the-art methods.

Knowledge base (KB) completion adds new facts to a KB by making inferences from existing facts, for example by inferring with high likelihood nationality(X,Y) from bornIn(X,Y). Most previous methods infer simple one-hop relational synonyms like this, or use as evidence a multi-hop relational path treated as an atomic feature, like bornIn(X,Z)→ containedIn(Z,Y). This paper presents an approach that reasons about conjunctions of multi-hop relations non-atomically, composing the implications of a path using a recurrent neural network (RNN) that takes as inputs vector embeddings of the binary relation in the path. Not only does this allow us to generalize to paths unseen at training time, but also, with a single high-capacity RNN, to predict new relation types not seen when the compositional model was trained (zero-shot learning). We assemble a new dataset of over 52M relational triples, and show that our method improves over a traditional classifier by 11%, and a method leveraging pre-trained embeddings by 7%.

We present paired learning and inference algorithms for significantly reducing computation and increasing speed of the vector dot products in the classifiers that are at the heart of many NLP components. This is accomplished by partitioning the features into a sequence of templates which are ordered such that high confidence can often be reached using only a small fraction of all features. Parameter estimation is arranged to maximize accuracy and early confidence in this sequence. Our approach is simpler and better suited to NLP than other related cascade methods. We present experiments in left-to-right part-of-speech tagging, named entity recognition, and transition-based dependency parsing. On the typical benchmarking datasets we can preserve POS tagging accuracy above 97% and parsing LAS above 88.5% both with over a five-fold reduction in run-time, and NER F1 above 88 with more than 2x increase in speed.

Vector space representation of words has been widely used to capture fine-grained linguistic regularities, and proven to be successful in various natural language processing tasks in recent years. However, existing models for learning word representations focus on either syntagmatic or paradigmatic relations alone. In this paper, we argue that it is beneficial to jointly modeling both relations so that we can not only encode different types of linguistic properties in a unified way, but also boost the representation learning due to the mutual enhancement between these two types of relations. We propose two novel distributional models for word representation using both syntagmatic and paradigmatic relations via a joint training objective. The proposed models are trained on a public Wikipedia corpus, and the learned representations are evaluated on word analogy and word similarity tasks. The results demonstrate that the proposed models can perform significantly better than all the state-of-the-art baseline methods on both tasks.

Entity classification, like many other important problems in NLP, involves learning classifiers over sparse highdimensional feature spaces that result from the conjunction of elementary features of the entity mention and its context. In this paper we develop a low-rank regularization framework for training maxentropy models in such sparse conjunctive feature spaces. Our approach handles conjunctive feature spaces using matrices and induces an implicit low-dimensional representation via low-rank constraints. We show that when learning entity classifiers under minimal supervision, using a seed set, our approach is more effective in controlling model capacity than standard techniques for linear classifiers.

Online debate forums present a valuable opportunity for the understanding and modeling of dialogue. To understand these debates, a key challenge is inferring the stances of the participants, all of which are interrelated and dependent. While collectively modeling users’ stances has been shown to be effective (Walker et al., 2012c; Hasan and Ng, 2013), there are many modeling decisions whose ramifications are not well understood. To investigate these choices and their effects, we introduce a scalable unified probabilistic modeling framework for stance classification models that 1) are collective, 2) reason about disagreement, and 3) can model stance at either the author level or at the post level. We comprehensively evaluate the possible modeling choices on eight topics across two online debate corpora, finding accuracy improvements of up to 11.5 percentage points over a local classifier. Our results highlight the importance of making the correct modeling choices for online dialogues, and having a unified probabilistic modeling framework that makes this possible.

Contrasting meaning is a basic aspect of semantics. Recent word-embedding models based on distributional semantics hypothesis are known to be weak for modeling lexical contrast. We present in this paper the embedding models that achieve an F-score of 92% on the widely-used, publicly available dataset, the GRE “most contrasting word” questions (Mohammad et al., 2008). This is the highest performance seen so far on this dataset. Surprisingly at the first glance, unlike what was suggested in most previous work, where relatedness statistics learned from corpora is claimed to yield extra gains over lexicon-based models, we obtained our best result relying solely on lexical resources (Roget’s and WordNet)—corpora statistics did not lead to further improvement. However, this should not be simply taken as that distributional statistics is not useful. We examine several basic concerns in modeling contrasting meaning to provide detailed analysis, with the aim to shed some light on the future directions for this basic semantics modeling problem.

Word embeddings have recently gained considerable popularity for modeling words in different Natural Language Processing (NLP) tasks including semantic similarity measurement. However, notwithstanding their success, word embeddings are by their very nature unable to capture polysemy, as different meanings of a word are conflated into a single representation. In addition, their learning process usually relies on massive corpora only, preventing them from taking advantage of structured knowledge. We address both issues by proposing a multifaceted approach that transforms word embeddings to the sense level and leverages knowledge from a large semantic network for effective semantic similarity measurement. We evaluate our approach on word similarity and relational similarity frameworks, reporting state-of-the-art performance on multiple datasets.

This paper considers the problem of embedding Knowledge Graphs (KGs) consisting of entities and relations into lowdimensional vector spaces. Most of the existing methods perform this task based solely on observed facts. The only requirement is that the learned embeddings should be compatible within each individual fact. In this paper, aiming at further discovering the intrinsic geometric structure of the embedding space, we propose Semantically Smooth Embedding (SSE). The key idea of SSE is to take full advantage of additional semantic information and enforce the embedding space to be semantically smooth, i.e., entities belonging to the same semantic category will lie close to each other in the embedding space. Two manifold learning algorithms Laplacian Eigenmaps and Locally Linear Embedding are used to model the smoothness assumption. Both are formulated as geometrically based regularization terms to constrain the embedding task. We empirically evaluate SSE in two benchmark tasks of link prediction and triple classification, and achieve significant and consistent improvements over state-of-the-art methods. Furthermore, SSE is a general framework. The smoothness assumption can be imposed to a wide variety of embedding models, and it can also be constructed using other information besides entities’ semantic categories.

Massive open online courses (MOOCs) are redefining the education system and transcending boundaries posed by traditional courses. With the increase in popularity of online courses, there is a corresponding increase in the need to understand and interpret the communications of the course participants. Identifying topics or aspects of conversation and inferring sentiment in online course forum posts can enable instructor interventions to meet the needs of the students, rapidly address course-related issues, and increase student retention. Labeled aspect-sentiment data for MOOCs are expensive to obtain and may not be transferable between courses, suggesting the need for approaches that do not require labeled data. We develop a weakly supervised joint model for aspectsentiment in online courses, modeling the dependencies between various aspects and sentiment using a recently developed scalable class of statistical relational models called hinge-loss Markov random fields. We validate our models on posts sampled from twelve online courses, each containing an average of 10,000 posts, and demonstrate that jointly modeling aspect with sentiment improves the prediction accuracy for both aspect and sentiment.

We present MultiGranCNN, a general deep learning architecture for matching text chunks. MultiGranCNN supports multigranular comparability of representations: shorter sequences in one chunk can be directly compared to longer sequences in the other chunk. MultiGranCNN also contains a flexible and modularized match feature component that is easily adaptable to different types of chunk matching. We demonstrate stateof-the-art performance of MultiGranCNN on clause coherence and paraphrase identification tasks.

The ability to map descriptions of scenes to 3D geometric representations has many applications in areas such as art, education, and robotics. However, prior work on the text to 3D scene generation task has used manually specified object categories and language that identifies them. We introduce a dataset of 3D scenes annotated with natural language descriptions and learn from this data how to ground textual descriptions to physical objects. Our method successfully grounds a variety of lexical terms to concrete referents, and we show quantitatively that our method improves 3D scene generation over previous work using purely rule-based methods. We evaluate the fidelity and plausibility of 3D scenes generated with our grounding approach through human judgments. To ease evaluation on this task, we also introduce an automated metric that strongly correlates with human judgments.

The Visual Dependency Representation (VDR) is an explicit model of the spatial relationships between objects in an image. In this paper we present an approach to training a VDR Parsing Model without the extensive human supervision used in previous work. Our approach is to find the objects mentioned in a given description using a state-of-the-art object detector, and to use successful detections to produce training data. The description of an unseen image is produced by first predicting its VDR over automatically detected objects, and then generating the text with a template-based generation model using the predicted VDR. The performance of our approach is comparable to a state-of-the-art multimodal deep neural network in images depicting actions.

We present a three-pronged approach to improving Statistical Machine Translation (SMT), building on recent success in the application of neural networks to SMT. First, we propose new features based on neural networks to model various non-local translation phenomena. Second, we augment the architecture of the neural network with tensor layers that capture important higher-order interaction among the network units. Third, we apply multitask learning to estimate the neural network parameters jointly. Each of our proposed methods results in significant improvements that are complementary. The overall improvement is +2.7 and +1.8 BLEU points for Arabic-English and Chinese-English translation over a state-of-the-art system that already includes neural network features.

The recently proposed neural network joint model (NNJM) (Devlin et al., 2014) augments the n-gram target language model with a heuristically chosen source context window, achieving state-of-the-art performance in SMT. In this paper, we give a more systematic treatment by summarizing the relevant source information through a convolutional architecture guided by the target information. With different guiding signals during decoding, our specifically designed convolution+gating architectures can pinpoint the parts of a source sentence that are relevant to predicting a target word, and fuse them with the context of entire source sentence to form a unified representation. This representation, together with target language words, are fed to a deep neural network (DNN) to form a stronger NNJM. Experiments on two NIST Chinese-English translation tasks show that the proposed model can achieve significant improvements over the previous NNJM by up to +1.08 BLEU points on average

Neural Machine Translation (NMT) is a new approach to machine translation that has shown promising results that are comparable to traditional approaches. A significant weakness in conventional NMT systems is their inability to correctly translate very rare words: end-to-end NMTs tend to have relatively small vocabularies with a single unk symbol that represents every possible out-of-vocabulary (OOV) word. In this paper, we propose and implement an effective technique to address this problem. We train an NMT system on data that is augmented by the output of a word alignment algorithm, allowing the NMT system to emit, for each OOV word in the target sentence, the position of its corresponding word in the source sentence. This information is later utilized in a post-processing step that translates every OOV word using a dictionary. Our experiments on the WMT’14 English to French translation task show that this method provides a substantial improvement of up to 2.8 BLEU points over an equivalent NMT system that does not use this technique. With 37.5 BLEU points, our NMT system is the first to surpass the best result achieved on a WMT’14 contest task.

Neural machine translation, a recently proposed approach to machine translation based purely on neural networks, has shown promising results compared to the existing approaches such as phrase-based statistical machine translation. Despite its recent success, neural machine translation has its limitation in handling a larger vocabulary, as training complexity as well as decoding complexity increase proportionally to the number of target words. In this paper, we propose a method based on importance sampling that allows us to use a very large target vocabulary without increasing training complexity. We show that decoding can be efficiently done even with the model having a very large target vocabulary by selecting only a small subset of the whole target vocabulary. The models trained by the proposed approach are empirically found to outperform the baseline models with a small vocabulary as well as the LSTM-based neural machine translation models. Furthermore, when we use the ensemble of a few models with very large target vocabularies, we achieve the state-of-the-art translation performance (measured by BLEU) on the English!German translation and almost as high performance as state-of-the-art English!French translation system.