ACL 2015 论文列表

We present a novel solution to improve the performance of Chinese word segmentation (CWS) using a synthetic word parser. The parser analyses the internal structure of words, and attempts to convert out-of-vocabulary words (OOVs) into in-vocabulary fine-grained sub-words. We propose a pipeline CWS system that first predicts this fine-grained segmentation, then chunks the output to reconstruct the original word segmentation standard. We achieve competitive results on the PKU and MSR datasets, with substantial improvements in OOV recall.

We define a dynamic oracle for the Covington non-projective dependency parser. This is not only the first dynamic oracle that supports arbitrary non-projectivity, but also considerably more efficient (O(n)) than the only existing oracle with restricted non-projectivity support. Experiments show that training with the dynamic oracle significantly improves parsing accuracy over the static oracle baseline on a wide range of treebanks.

Recent work on supertagging using a feedforward neural network achieved significant improvements for CCG supertagging and parsing (Lewis and Steedman, 2014). However, their architecture is limited to considering local contexts and does not naturally model sequences of arbitrary length. In this paper, we show how directly capturing sequence information using a recurrent neural network leads to further accuracy improvements for both supertagging (up to 1.9%) and parsing (up to 1% F1), on CCGBank, Wikipedia and biomedical text.

We present KLcpos3 , a language similarity measure based on Kullback-Leibler divergence of coarse part-of-speech tag trigram distributions in tagged corpora. It has been designed for multilingual delexicalized parsing, both for source treebank selection in single-source parser transfer, and for source treebank weighting in multi-source transfer. In the selection task, KLcpos3 identifies the best source treebank in 8 out of 18 cases. In the weighting task, it brings +4.5% UAS absolute, compared to unweighted parse tree combination.

We present a novel scheme for wordbased Japanese typed dependency parser which integrates syntactic structure analysis and grammatical function analysis such as predicate-argument structure analysis. Compared to bunsetsu-based dependency parsing, which is predominantly used in Japanese NLP, it provides a natural way of extracting syntactic constituents, which is useful for downstream applications such as statistical machine translation. It also makes it possible to jointly decide dependency and predicate-argument structure, which is usually implemented as two separate steps. We convert an existing treebank to the new dependency scheme and report parsing results as a baseline for future research. We achieved a better accuracy for assigning function labels than a predicate-argument structure analyzer by using grammatical functions as dependency label.

Multi-modal semantics has relied on feature norms or raw image data for perceptual input. In this paper we examine grounding semantic representations in olfactory (smell) data, through the construction of a novel bag of chemical compounds model. We use standard evaluations for multi-modal semantics, including measuring conceptual similarity and cross-modal zero-shot learning. To our knowledge, this is the first work to evaluate semantic similarity on representations grounded in olfactory data.

This paper addresses a novel task of semantically analyzing the comparative constructions inherent in attributive superlative expressions against structured knowledge bases (KBs). The task can be defined in two-fold: first, selecting the comparison dimension against a KB, on which the involved items are compared; and second, determining the ranking order, in which the items are ranked (ascending or descending). We exploit Wikipedia and Freebase to collect training data in an unsupervised manner, where a neural network model is then learnt to select, from Freebase predicates, the most appropriate comparison dimension for a given superlative expression, and further determine its ranking order heuristically. Experimental results show that it is possible to learn from coarsely obtained training data to semantically characterize the comparative constructions involved in attributive superlative expressions.

We consider the task of identifying and labeling the semantic arguments of a predicate that evokes a FrameNet frame. This task is challenging because there are only a few thousand fully annotated sentences for supervised training. Our approach augments an existing model with features derived from FrameNet and PropBank and with partially annotated exemplars from FrameNet. We observe a 4% absolute increase in F1 versus the original model.

A defining symptom of autism spectrum disorder (ASD) is the presence of restricted and repetitive activities and interests, which can surface in language as a perseverative focus on idiosyncratic topics. In this paper, we use semantic similarity measures to identify such idiosyncratic topics in narratives produced by children with and without ASD. We find that neurotypical children tend to use the same words and semantic concepts when retelling the same narrative, while children with ASD, even when producing accurate retellings, use different words and concepts relative not only to neurotypical children but also to other children with ASD. Our results indicate that children with ASD not only stray from the target topic but do so in idiosyncratic ways according to their own restricted interests.

With massively parallel corpora of hundreds or thousands of translations of the same text, it is possible to automatically perform typological studies of language structure using very large languag ...

Automatic resolution of Crossword Puzzles (CPs) heavily depends on the quality of the answer candidate lists produced by a retrieval system for each clue of the puzzle grid. Previous work has shown that such lists can be generated using Information Retrieval (IR) search algorithms applied to the databases containing previously solved CPs and reranked with tree kernels (TKs) applied to a syntactic tree representation of the clues. In this paper, we create a labelled dataset of 2 million clues on which we apply an innovative Distributional Neural Network (DNN) for reranking clue pairs. Our DNN is computationally efficient and can thus take advantage of such large datasets showing a large improvement over the TK approach, when the latter uses small training data. In contrast, when data is scarce, TKs outperform DNNs.

In this paper, we apply the concept of pretraining to hidden-unit conditional random fields (HUCRFs) to enable learning on unlabeled data. We present a simple yet effective pre-training technique that learns to associate words with their clusters, which are obtained in an unsupervised manner. The learned parameters are then used to initialize the supervised learning process. We also propose a word clustering technique based on canonical correlation analysis (CCA) that is sensitive to multiple word senses, to further improve the accuracy within the proposed framework. We report consistent gains over standard conditional random fields (CRFs) and HUCRFs without pre-training in semantic tagging, named entity recognition (NER), and part-of-speech (POS) tagging tasks, which could indicate the task independent nature of the proposed technique.

Log-bilinear language models such as SkipGram and GloVe have been proven to capture high quality syntactic and semantic relationships between words in a vector space. We revisit the relationship between SkipGram and GloVe models from a machine learning viewpoint, and show that these two methods are easily merged into a unified form. Then, by using the unified form, we extract the factors of the configurations that they use differently. We also empirically investigate which factor is responsible for the performance difference often observed in widely examined word similarity and analogy tasks.

Text regression has traditionally been tackled using linear models. Here we present a non-linear method based on a deep convolutional neural network. We show that despite having millions of parameters, this model can be trained on only a thousand documents, resulting in a 40% relative improvement over sparse linear models, the previous state of the art. Further, this method is flexible allowing for easy incorporation of side information such as document meta-data. Finally we present a novel technique for interpreting the effect of different text inputs on this complex non-linear model.

In sentence modeling and classification, convolutional neural network approaches have recently achieved state-of-the-art results, but all such efforts process word vectors sequentially and neglect long-distance dependencies. To exploit both deep learning and linguistic structures, we propose a tree-based convolutional neural network model which exploit various long-distance relationships between words. Our model improves the sequential baselines on all three sentiment and question classification tasks, and achieves the highest published accuracy on TREC.

We present a simple yet effective unsupervised domain adaptation method that can be generally applied for different NLP tasks. Our method uses unlabeled target domain instances to induce a set of instance similarity features. These features are then combined with the original features to represent labeled source domain instances. Using three NLP tasks, we show that our method consistently outperforms a few baselines, including SCL, an existing general unsupervised domain adaptation method widely used in NLP. More importantly, our method is very easy to implement and incurs much less computational cost than SCL.

Replicated Softmax model, a well-known undirected topic model, is powerful in extracting semantic representations of documents. Traditional learning strategies such as Contrastive Divergence are very inefficient. This paper provides a novel estimator to speed up the learning based on Noise Contrastive Estimate, extended for documents of variant lengths and weighted inputs. Experiments on two benchmarks show that the new estimator achieves great learning efficiency and high accuracy on document retrieval and classification.

Downstream processing of machine translation (MT) output promises to be a solution to improve translation quality, especially when the MT system’s internal decoding process is not accessible. Both rule-based and statistical automatic postediting (APE) methods have been proposed over the years, but with contrasting results. A missing aspect in previous evaluations is the assessment of different methods: i) under comparable conditions, and ii) on different language pairs featuring variable levels of MT quality. Focusing on statistical APE methods (more portable across languages), we propose the first systematic analysis of two approaches. To understand their potential, we compare them in the same conditions over six language pairs having English as source. Our results evidence consistent improvements on all language pairs, a relation between the extent of the gain and MT output quality, slight but statistically significant performance differences between the two methods, and their possible complementarity.

Precisely evaluating the quality of a translation against human references is a challenging task due to the flexible word ordering of a sentence and the existence of a large number of synonyms for words. This paper proposes to evaluate translations with distributed representations of words and sentences. We study several metrics based on word and sentence representations and their combination. Experiments on the WMT metric task shows that the metric based on the combined representations achieves the best performance, outperforming the state-of-the-art translation metrics by a large margin. In particular, training the distributed representations only needs a reasonable amount of monolingual, unlabeled data that is not necessary drawn from the test domain.