COLING 2016 论文列表

We present Ambient Search, an open source system for displaying and retrieving relevant documents in real time for speech input. The system works ambiently, that is, it unobstructively listens to speech streams in the background, identifies keywords and keyphrases for query construction and continuously serves relevant documents from its index. Query terms are ranked with Word2Vec and TF-IDF and are continuously updated to allow for ongoing querying of a document collection. The retrieved documents, in our case Wikipedia articles, are visualized in real time in a browser interface. Our evaluation shows that Ambient Search compares favorably to another implicit information retrieval system on speech streams. Furthermore, we extrinsically evaluate multiword keyphrase generation, showing positive impact for manual transcriptions.

We introduce the task of detecting cross-lingual marketing blunders, which occur if a trade name resembles an inappropriate or negatively connotated word in a target language. To this end, we suggest a formal task definition and a semi-automatic method based the propagation of pragmatic labels from Wiktionary across sense-disambiguated translations. Our final tool assists users by providing clues for problematic names in any language, which we simulate in two experiments on detecting previously occurred marketing blunders and identifying relevant clues for established international brands. We conclude the paper with a suggested research roadmap for this new task. To initiate further research, we publish our online demo along with the source code and data at http://uby.ukp.informatik.tu-darmstadt.de/blunder/.

Training good word embeddings requires large amounts of data. Out-of-vocabulary words will still be encountered at test-time, leaving these words without embeddings. To overcome this lack of embeddings for rare words, existing methods leverage morphological features to generate embeddings. While the existing methods use computationally-intensive rule-based (Soricut and Och, 2015) or tool-based (Botha and Blunsom, 2014) morphological analysis to generate embeddings, our system applies a computationally-simpler sub-word search on words that have existing embeddings. Embeddings of the sub-word search results are then combined using string similarity functions to generate rare word embeddings. We augmented pre-trained word embeddings with these novel embeddings and evaluated on a rare word similarity task, obtaining up to 3 times improvement in correlation over the original set of embeddings. Applying our technique to embeddings trained on larger datasets led to on-par performance with the existing state-of-the-art for this task. Additionally, while analysing augmented embeddings in a log-bilinear language model, we observed up to 50% reduction in rare word perplexity in comparison to other more complex language models.

In many applications such as personal digital assistants, there is a constant need for new domains to increase the system’s coverage of user queries. A conventional approach is to learn a separate model every time a new domain is introduced. This approach is slow, inefficient, and a bottleneck for scaling to a large number of domains. In this paper, we introduce a framework that allows us to have a single model that can handle all domains: including unknown domains that may be created in the future as long as they are covered in the master schema. The key idea is to remove the need for distinguishing domains by explicitly predicting the schema of queries. Given permitted schema of a query, we perform constrained decoding on a lattice of slot sequences allowed under the schema. The proposed model achieves competitive and often superior performance over the conventional model trained separately per domain.

This work proposes a new confidence measure for evaluating text-to-speech alignment systems outputs, which is a key component for many applications, such as semi-automatic corpus anonymization, lips syncing, film dubbing, corpus preparation for speech synthesis and speech recognition acoustic models training. This confidence measure exploits deep neural networks that are trained on large corpora without direct supervision. It is evaluated on an open-source spontaneous speech corpus and outperforms a confidence score derived from a state-of-the-art text-to-speech aligner. We further show that this confidence measure can be used to fine-tune the output of this aligner and improve the quality of the resulting alignment.

Natural language generation (NLG) is an important component of question answering(QA) systems which has a significant impact on system quality. Most tranditional QA systems based on templates or rules tend to generate rigid and stylised responses without the natural variation of human language. Furthermore, such methods need an amount of work to generate the templates or rules. To address this problem, we propose a Context-Aware LSTM model for NLG. The model is completely driven by data without manual designed templates or rules. In addition, the context information, including the question to be answered, semantic values to be addressed in the response, and the dialogue act type during interaction, are well approached in the neural network model, which enables the model to produce variant and informative responses. The quantitative evaluation and human evaluation show that CA-LSTM obtains state-of-the-art performance.

An interactive Question Answering (QA) system frequently encounters non-sentential (incomplete) questions. These non-sentential questions may not make sense to the system when a user asks them without the context of conversation. The system thus needs to take into account the conversation context to process the question. In this work, we present a recurrent neural network (RNN) based encoder decoder network that can generate a complete (intended) question, given an incomplete question and conversation context. RNN encoder decoder networks have been show to work well when trained on a parallel corpus with millions of sentences, however it is extremely hard to obtain conversation data of this magnitude. We therefore propose to decompose the original problem into two separate simplified problems where each problem focuses on an abstraction. Specifically, we train a semantic sequence model to learn semantic patterns, and a syntactic sequence model to learn linguistic patterns. We further combine syntactic and semantic sequence models to generate an ensemble model. Our model achieves a BLEU score of 30.15 as compared to 18.54 using a standard RNN encoder decoder model.

In this study, we applied a deep LSTM structure to classify dialogue acts (DAs) in open-domain conversations. We found that the word embeddings parameters, dropout regularization, decay rate and number of layers are the parameters that have the largest effect on the final system accuracy. Using the findings of these experiments, we trained a deep LSTM network that outperforms the state-of-the-art on the Switchboard corpus by 3.11%, and MRDA by 2.2%.

Identifying dialogue acts and dialogue modes during tutorial interactions is an extremely crucial sub-step in understanding patterns of effective tutor-tutee interactions. In this work, we develop a novel joint inference method that labels each utterance in a tutoring dialogue session with a dialogue act and a specific mode from a set of pre-defined dialogue acts and modes, respectively. Specifically, we develop our joint model using Markov Logic Networks (MLNs), a framework that combines first-order logic with probabilities, and is thus capable of representing complex, uncertain knowledge. We define first-order formulas in our MLN that encode the inter-dependencies between dialogue modes and more fine-grained dialogue actions. We then use a joint inference to jointly label the modes as well as the dialogue acts in an utterance. We compare our system against a pipeline system based on SVMs on a real-world dataset with tutoring sessions of over 500 students. Our results show that the joint inference system is far more effective than the pipeline system in mode detection, and improves over the performance of the pipeline system by about 6 points in F1 score. The joint inference system also performs much better than the pipeline system in the context of labeling modes that highlight important pedagogical steps in tutoring.

While automatic response generation for building chatbot systems has drawn a lot of attention recently, there is limited understanding on when we need to consider the linguistic context of an input text in the generation process. The task is challenging, as messages in a conversational environment are short and informal, and evidence that can indicate a message is context dependent is scarce. After a study of social conversation data crawled from the web, we observed that some characteristics estimated from the responses of messages are discriminative for identifying context dependent messages. With the characteristics as weak supervision, we propose using a Long Short Term Memory (LSTM) network to learn a classifier. Our method carries out text representation and classifier learning in a unified framework. Experimental results show that the proposed method can significantly outperform baseline methods on accuracy of classification.

Music information retrieval has emerged as a mainstream research area in the past two decades. Experiments on music mood classification have been performed mainly on Western music based on audio, lyrics and a combination of both. Unfortunately, due to the scarcity of digitalized resources, Indian music fares poorly in music mood retrieval research. In this paper, we identified the mood taxonomy and prepared multimodal mood annotated datasets for Hindi and Western songs. We identified important audio and lyric features using correlation based feature selection technique. Finally, we developed mood classification systems using Support Vector Machines and Feed Forward Neural Networks based on the features collected from audio, lyrics, and a combination of both. The best performing multimodal systems achieved F-measures of 75.1 and 83.5 for classifying the moods of the Hindi and Western songs respectively using Feed Forward Neural Networks. A comparative analysis indicates that the selected features work well for mood classification of the Western songs and produces better results as compared to the mood classification systems for Hindi songs.

In this paper we focus on the problem of dialog act (DA) labelling. This problem has recently attracted a lot of attention as it is an important sub-part of an automatic question answering system, which is currently in great demand. Traditional methods tend to see this problem as a sequence labelling task and deals with it by applying classifiers with rich features. Most of the current neural network models still omit the sequential information in the conversation. Henceforth, we apply a novel multi-level gated recurrent neural network (GRNN) with non-textual information to predict the DA tag. Our model not only utilizes textual information, but also makes use of non-textual and contextual information. In comparison, our model has shown significant improvement over previous works on Switchboard Dialog Act (SWDA) task by over 6%.

This study proposes a computational model of the discourse segments in lyrics to understand and to model the structure of lyrics. To test our hypothesis that discourse segmentations in lyrics strongly correlate with repeated patterns, we conduct the first large-scale corpus study on discourse segments in lyrics. Next, we propose the task to automatically identify segment boundaries in lyrics and train a logistic regression model for the task with the repeated pattern and textual features. The results of our empirical experiments illustrate the significance of capturing repeated patterns in predicting the boundaries of discourse segments in lyrics.

Cross document event coreference (CDEC) is an important task that aims at aggregating event-related information across multiple documents. We revisit the evaluation for CDEC, and discover that past works have adopted different, often inconsistent, evaluation settings, which either overlook certain mistakes in coreference decisions, or make assumptions that simplify the coreference task considerably. We suggest a new evaluation methodology which overcomes these limitations, and allows for an accurate assessment of CDEC systems. Our new evaluation setting better reflects the corpus-wide information aggregation ability of CDEC systems by separating event-coreference decisions made across documents from those made within a document. In addition, we suggest a better baseline for the task and semi-automatically identify several inconsistent annotations in the evaluation dataset.

Determining the relative order of events and times described in text is an important problem in natural language processing. It is also a difficult one: general state-of-the-art performance has been stuck at a relatively low ceiling for years. We investigate the representation of temporal relations, and empirically evaluate the effect that various temporal relation representations have on machine learning performance. While machine learning performance decreases with increased representational expressiveness, not all representation simplifications have equal impact.

This paper introduces a novel method for measuring non-cooperation in dialogue. The key idea is that linguistic non-cooperation can be measured in terms of the extent to which dialogue participants deviate from conventions regarding the proper introduction and discharging of conversational obligations (e.g., the obligation to respond to a question). Previous work on non cooperation has focused mainly on non-linguistic task-related non-cooperation or modelled non-cooperation in terms of special rules describing non-cooperative behaviours. In contrast, we start from rules for normal/correct dialogue behaviour - i.e., a dialogue game - which in principle can be derived from a corpus of cooperative dialogues, and provide a quantitative measure for the degree to which participants comply with these rules. We evaluated the model on a corpus of political interviews, with encouraging results. The model predicts accurately the degree of cooperation for one of the two dialogue game roles (interviewer) and also the relative cooperation for both roles (i.e., which interlocutor in the conversation was most cooperative). Being able to measure cooperation has applications in many areas from the analysis - manual, semi and fully automatic - of natural language interactions to human-like virtual personal assistants, tutoring agents, sophisticated dialogue systems, and role-playing virtual humans.

For the task of implicit discourse relation recognition, traditional models utilizing manual features can suffer from data sparsity problem. Neural models provide a solution with distributed representations, which could encode the latent semantic information, and are suitable for recognizing semantic relations between argument pairs. However, conventional vector representations usually adopt embeddings at the word level and cannot well handle the rare word problem without carefully considering morphological information at character level. Moreover, embeddings are assigned to individual words independently, which lacks of the crucial contextual information. This paper proposes a neural model utilizing context-aware character-enhanced embeddings to alleviate the drawbacks of the current word level representation. Our experiments show that the enhanced embeddings work well and the proposed model obtains state-of-the-art results.

We experiment with different ways of training LSTM networks to predict RST discourse trees. The main challenge for RST discourse parsing is the limited amounts of training data. We combat this by regularizing our models using task supervision from related tasks as well as alternative views on discourse structures. We show that a simple LSTM sequential discourse parser takes advantage of this multi-view and multi-task framework with 12-15% error reductions over our baseline (depending on the metric) and results that rival more complex state-of-the-art parsers.

In this paper, we investigate four important issues together for explicit discourse relation labelling in Chinese texts: (1) discourse connective extraction, (2) linking ambiguity resolution, (3) relation type disambiguation, and (4) argument boundary identification. In a pipelined Chinese discourse parser, we identify potential connective candidates by string matching, eliminate non-discourse usages from them with a binary classifier, resolve linking ambiguities among connective components by ranking, disambiguate relation types by a multiway classifier, and determine the argument boundaries by conditional random fields. The experiments on Chinese Discourse Treebank show that the F1 scores of 0.7506, 0.7693, 0.7458, and 0.3134 are achieved for discourse usage disambiguation, linking disambiguation, relation type disambiguation, and argument boundary identification, respectively, in a pipelined Chinese discourse parser.