Numerous online conversations are produced on a daily basis, resulting in a pressing need to conversation understanding. As a basis to structure a discussion, we identify the responding relations in the conversation discourse, which link response utterances to their initiations. To figure out who responded to whom, here we explore how the consistency of topic contents and dependency of discourse roles indicate such interactions, whereas most prior work ignore the effects of latent factors underlying word occurrences. We propose a model to learn latent topics and discourse in word distributions, and predict pairwise initiation-response links via exploiting topic consistency and discourse dependency. Experimental results on both English and Chinese conversations show that our model significantly outperforms the previous state of the arts, such as 79 vs. 73 MRR on Chinese customer service dialogues. We further probe into our outputs and shed light on how topics and discourse indicate conversational user interactions.
Sub-character components of Chinese characters carry important semantic information, and recent studies have shown that utilizing this information can improve performance on core semantic tasks. In this paper, we hypothesize that in addition to semantic information, sub-character components may also carry emotional information, and that utilizing it should improve performance on sentiment analysis tasks. We conduct a series of experiments on four Chinese sentiment data sets and show that we can significantly improve the performance in various tasks over that of a character-level embeddings baseline. We then focus on qualitatively assessing multiple examples and trying to explain how the sub-character components affect the results in each case.
Answer extraction is the most important part of a chinese web-based question answering system. In order to enhance the robustness and adaptability of answer extraction to new domains and eliminate the influence of the incomplete and noisy search snippets, we propose two new answer exraction methods. We utilize text patterns to generate Part-of-Speech (POS) patterns. In addition, a method is proposed to construct a POS tree by using these POS patterns. The POS tree is useful to candidate answer extraction of web-based question answering. To retrieve a efficient POS tree, the similarities between questions are used to select the question-answer pairs whose questions are similar to the unanswered question. Then, the POS tree is improved based on these question-answer pairs. In order to rank these candidate answers, the weights of the leaf nodes of the POS tree are calculated using a heuristic method. Moreover, the Genetic Algorithm (GA) is used to train the weights. The experimental results of 10-fold crossvalidation show that the weighted POS tree trained by GA can improve the accuracy of answer extraction.
For practical chatbots, one of the essential factor for improving user experience is the capability of customizing the talking style of the agents, that is, to make chatbots provide responses meeting users’ preference on language styles, topics, etc. To address this issue, this paper proposes to incorporate linguistic biases, which implicitly involved in the conversation corpora generated by human groups in the Social Network Services (SNS), into the encoder-decoder based response generator. By attaching a specially designed neural component to dynamically control the impact of linguistic biases in response generation, a Group Linguistic Bias Aware Neural Response Generation (GLBA-NRG) model is eventually presented. The experimental results on the dataset from the Chinese SNS show that the proposed architecture outperforms the current response generating models by producing both meaningful and vivid responses with customized styles.
The state-of-the-art Chinese word segmentation systems obtain high accuracy in domains like newswire but suffer a significant performance degradation when they are used in other domains such as patents and literature. In this paper, we propose a neural domain adaptation approach which works through cross-domain embeddings and uses unlabeled target domain data to improve the cross-domain performance. Experiment results show that the proposed method achieves competitive performance with previous Chinese word segmentation domain adaptation methods.
Terms extensively exist in specific domains, and term translation plays a critical role in domain-specific machine translation (MT) tasks. However, it’s a challenging task to translate them correctly for the huge number of pre-existing terms and the endless new terms. To achieve better term translation quality, it is necessary to inject external term knowledge into the underlying MT system. Fortunately, there are plenty of term translation knowledge in parenthetical sentences on the Internet. In this paper, we propose a simple, straightforward and effective framework to improve term translation by learning from parenthetical sentences. This framework includes: (1) a focused web crawler; (2) a parenthetical sentence filter, acquiring parenthetical sentences including bilingual term pairs; (3) a term translation knowledge extractor, extracting bilingual term translation candidates; (4) a probability learner, generating the term translation table for MT decoders. The extensive experiments demonstrate that our proposed framework significantly improves the translation quality of terms and sentences.
A central problem in research on automatic proficiency scoring is to differentiate the variability between and within groups of standard and non-standard speakers. Along with the effort to improve the robustness of techniques and models, we can also select test sentences that are more reliable for measuring the between-group variability. This study demonstrated that the performance of an automatic scoring system could be significantly improved by excluding “bad” sentences from the scoring procedure. The experiments on a dataset of Putonghua Shuiping Ceshi (Mandarin proficiency test) showed that, compared to all available sentences, using only best-performed sentences improved the speaker-level correlation between human and automatic scores from r = .640 to r = .824.
Named entity recognition (NER) plays an important role in the NLP literature. The traditional methods tend to employ large annotated corpus to achieve a high performance. Different with many semi-supervised learning models for NER task, in this paper, we employ the graph-based semi-supervised learning (GBSSL) method to utilize the freely available unlabeled data. The experiment shows that the unlabeled corpus can enhance the state-of-theart conditional random field (CRF) learning model and has potential to improve the tagging accuracy even though the margin is a little weak and not satisfying in current experiments.
This paper presents our Chinese microblog sentiment classification (CMSC) system in the Topic-Based Chinese Message Polarity Classification task of SIGHAN-8 Bake-Off. Given a message from Chinese Weibo platform and a topic, our system is designed to classify whether the message is of positive, negative, or neutral sentiment towards the given topic. Due to the difficulties like the out-ofvocabulary Internet words and emoticons, polarity classification of Chinese microblogs is still an open problem today. In our system, Maximum Entropy (MaxEnt) is employed, which is a discriminative model that directly models the class posteriors, allowing them to incorporate a rich set of features. Moreover, oversampling approach is used to hand the unbalance problem. Evaluation results demonstrate the utility of our system, showing an accuracy of 66.4% for restricted resource and 66.6% for unrestricted resource.
In this study, an automatic classification method based on the sentiment polarity of text is proposed. This method uses two sentiment dictionaries from different sources: the Chinese sentiment dictionary CSWN that integrates Chinese WordNet with SentiWordNet, and the sentiment dictionary obtained from a training corpus labeled with sentiment polarities. In this study, the sentiment polarity of text is analyzed using these two dictionaries, a mixed-rule approach, and a statistics-based prediction model. The proposed method is used to analyze a test corpus provided by the Topic-Based Chinese Message Polarity Classification task of SIGHAN-8, and the F1measure value is tested at 0.62.
Increased interest in China from foreigners has led to a corresponding interest in the study of Chinese.However, the learning of Chinese by non-native speakers will encounter many difficulties, Chinese spelling check techniques for Chinese as a Foreign Language(CFL) learners is highly desirable.This paper presents our work on the SIGHAN-2015 Chinese Spelling Check task.The task focuses on spelling checking on Chinese essays written by CFL learners.We propose a unified framework called HANSpeller++ based on our previous HANSpeller for Chinese spelling correction.The framework consists of candidate generating,candidates re-ranking and final global decision making.Experiments show good performance on the test data of the task.
This paper describes details of NTOU Chinese spelling check system in SIGHAN-8 Bakeoff.Besides the basic architecture of the previous system participating in last two CSC tasks, three new preference rules were proposed to deal with Simplified Chinese characters, variants, sentence-final particles, and DE-particles.A new sentence likelihood function was proposed based on frequencies of space-removed version of Google n-gram datasets.Two formal runs were submitted where the best one was created by the system using Google n-gram frequency information.
The manual Chinese word segmentation dataset WordSegCHC 1.0 which was built by eight crowdsourcing tasks conducted on the Crowdflower platform contains the manual word segmentation data of 152 Chinese sentences whose length ranges from 20 to 46 characters without punctuations.All the sentences received 200 segmentation responses in their corresponding crowdsourcing tasks and the numbers of valid response of them range from 123 to 143 (each sentence was segmented by more than 120 subjects).We also proposed an evaluation method called manual segmentation error rate (M SER) to evaluate the dataset; the M SER of the dataset is proved to be very low which indicates reliable data quality.In this work, we applied the crowdsourcing method to Chinese word segmentation task and the results confirmed again that the crowdsourcing method is a promising tool for linguistic data collection; the framework of crowdsourcing linguistic data collection used in this work can be reused in similar tasks; the resultant dataset filled a gap in Chinese language resources to the best of our knowledge, and it has potential applications in the research of word intuition of Chinese speakers and Chinese language processing.
We propose a linguistically driven approach to represent discourse relations in Chinese text as sequences. We observe that certain surface characteristics of Chinese texts, such as the order of clauses, are overt markers of discourse structures, yet existing annotation proposals adapted from formalism constructed for English do not fully incorporate these characteristics. We present an annotated resource consisting of 325 articles in the Chinese Treebank. In addition, using this annotation, we introduce a discourse chunker based on a cascade of classifiers and report 70% top-level discourse sense accuracy.
In this paper, we describe our system for the topic-based Chinese message polarity classification in SIGHAN 8 Task 2. Our system integrates two SVM classifiers which consist of LinearSVC and LibSVM to train the classification model and predict the results of Chinese message polarity in the restricted resource and the unrestricted resource, respectively.In order to assure our feature engineering effort on the task, we use some feature selection methods, such as LDA, word2vec, and sentiment lexicons including DLUT emotion ontology and NTUSD.Our system achieves the overall F1 score of 74.88% in the restricted evaluation and 74.43% in the unrestricted evaluation.
In this paper, we focus on topic-based microblog sentiment classification task that classify the microblog's sentiment polarities toward a specific topic.Most of the existing approaches for sentiment analysis usually adopt the target-independent strategy, which may assign irrelevant sentiments to the given topic.In this paper, we leverage the non-negative matrix factorization to get the relevant topic words and then further incorporate target-dependent features for topic-based microblog sentiment classification.According to the experiment results, our system (NDMSCS) has achieved a good performance in the SIGHAN 8 Task 2.
This paper presents our system in the Chinese spelling check (CSC) task of SIGHAN-8 Bake-Off. Given a sentence, our systems are designed to detect and correct the spelling error. As we know, CSC is still a hot topic today and it is an open problem yet. N-gram language modeling (LM) is widely used in CSC, since its simplicity and power. We present a model based on joint bi-gram and trigram LM and Chinese word segmentation. Besides, we apply dynamic programming to increase efficiency and employ smoothing technique to address the sparseness of the n-gram in training data. The evaluation results show the utility of our CSC system.
Previous researches have focused on analyzing emotion through monolingual text, when in fact bilingual or code-switching posts are also common in social media.Despite the important implications of code-switching for emotion analysis, existing automatic emotion extraction methods fail to accommodate for the code-switching content.In this paper, we propose a general framework to construct and analyze the code-switching emotional posts in social media.We first propose an annotation scheme to identify the emotions associated with the languages expressing them in a Chinese-English code-switching corpus.We then make some observations and generate statistics from the corpus to analyze the linguistic phenomena of code-switching texts in social media.Finally, we propose a multiple-classifier-based automatic detection approach to detect emotion in the codeswitching corpus for evaluating the effectiveness of both Chinese and English texts.
This paper presents an application of Chinese syntactic knowledge for semantic role labeling (SRL).Besides basic morphological information, syntactic structures are crucial in SRL.However, it is difficult to learn such information from limited, small-scale, manually annotated training data.Instead of manually increasing the size of annotated data, we use a large amount of automatically extracted syntactic knowledge to improve the performance of SRL.
This paper describes our system (MSIIP THU) used for Topic-Based Chinese Message Polarity Classification Task in SIGHAN-8. In our system, a lexiconbased classifier and a statistical machine learning-based classifier are built up, followed by a linear combination of these two models. The overall performance of the proposed framework ranks in the middle of all terms participating in the task.
