Search

Distributional Semantic Models (DSMs) established themselves as a standard for the representation of word and sentence meaning. However, DSMs provide quantitative measurement of how strongly two linguistic expressions are related, without being able to automatically classify different semantic relations. Hence the notion of semantic similarity is underspecified in DSMs. We introduce Evalution-MAN in this paper as an effort to address this underspecification problem. Following the EVALution 1.0 dataset for English, we present a dataset for evaluating DSMs on the task of the identification of semantic relations in Mandarin Chinese. Moreover, we test different types of word vectors on the automatic learning of these semantic relations, and we evaluate them both in a unsupervised and in a supervised setting, finding that distributional models tend, in general, to assign higher similarity scores to synonyms and that deep learning classifiers are the best performing ones in the identification of semantic relations.

Finding methods to represent multiple types of nodes in heterogeneous networks is both challenging and rewarding, as there is much less work in this area compared with that of homogeneous networks. In this paper, we propose a novel approach to learn node embedding for heterogeneous networks through a joint learning framework of both network links and text associated with nodes. A novel attention mechanism is also used to make good use of text extended through links to obtain much larger network context. Link embedding is first learned through a random-walk-based method to process multiple types of links. Text embedding is separately learned at both sentence level and document level to capture salient semantic information more comprehensively. Then, both types of embeddings are jointly fed into a hierarchical neural network model to learn node representation through mutual enhancement. The attention mechanism follows linked edges to obtain context of adjacent nodes to extend context for node representation. The evaluation on a link prediction task in a heterogeneous network data set shows that our method outperforms the current state-of-the-art method by 2.5%-5.0% in AUC values with p-value less than 10-9, indicating very significant improvement.

Inferring latent user preferences using both structured and unstructured data is an important social computing task. In this paper, we propose a user preference representation based on user activities embedded in unstructured data to better encode the homophily theory. The representation of an individual user is learned using a embedding based method to integrate latent user preferences in social media. The method has the ability to integrate a variety of user activities based cues from user comments, user social network (i.e; follower/followee connections) and user interested topics which are indicated by the topics a user has participated in. Experiments are conducted to evaluate the prediction of each user's favorite team as a part of user preferences in a dataset collected from the Hu-pu basketball discussion forum.1 Results clearly indicate that our proposed user representation outperforms other user representation baselines. Integrating user social network and user interested topics with user comments can improve the overall performance of user preference prediction.

The biggest challenges to rules-based approaches to Natural Language Processing (NLP) are the resources required to do an exhaustive search for rule-matching, and the decision to select the optimal rule when there are multiple possible matches. In this paper, we propose a novel approach named pattern-based rule disambiguation (PRD) to face these challenges. PRD helps to determine which rule is activated by a pattern when the pattern activates more than one rule. To tackle this task, we first collect and annotate the samples following the same pattern, but activating different rules; Then, we leverage the corpus to train a statistic classifier to disambiguate the pattern. This new approach is applied to the task of emotion cause detection, adopting a linguistic rule-drive paradigm which was the only one available for this task. The experimental results demonstrated the effectiveness of our PRD approach and offered a promising solution of the resolution of multiple-matched rules challenge for future NLP tasks.

Sentiment classification is now a hot research issue in the community of natural language processing and the bag-of-words based machine learning approach is the state-of-the-art for this task. However, one important phenomenon, called polarity shifting, remains unsolved in the bag-of-words model, which sometimes makes the machine learning approach fails. In this study, we aim to perform sentiment classification with full consideration of the polarity shifting phenomenon. First, we extract some detection rules for detecting polarity shifting of sentimental words from a corpus which consists of polarity-shifted sentences. Then, we use the detection rules to detect the polarity-shifted words in the testing data. Third, a novel term counting-based classifier is designed by fully considering those polarity-shifted words. Evaluation shows that the novel term counting-based classifier significantly improves the performance of sentiment analysis across five domains. Furthermore, when this classifier is combined with a machine-learning based classifier, the combined classifier yields better performance than either of them.

Cross-lingual sentiment classification aims to perform sentiment classification in a language (named as the target language) with the help of the resources from another language (named as the source language). Previous studies are prone to using all available data in the source language while using all data is observed to perform no better or even worse than using a partion of good data. In this paper, we propose a novel task called data quality controlling in the source language to select high quality samples from the source language. To tackle this task, we propose two kinds of data quality measurements: intra- and extra-quality measurements which are implemented with the certainty and similarity measurements respectively. The empirical studies demonstrate the effectiveness of the proposed approach to data quality controlling in the source language.

The intensive studies have shown that the information in the context of the mentions of a person is very helpful for personal name disambiguation. However, it is not easy to exact all mentions of a person in text. In this paper, we investigate various mentions of a person in English and Chinese, and find that different languages often emphasize different technologies to exact those mentions of a person. Based on this observation, we make an in-depth analysis of a Chinese personal name disambiguation corpus in terms of structures and meanings of Chinese personal names. Finally, we design two evaluation schemes to examine the impact of selecting personal names to Chinese personal name disambiguation. Our experiments show that deeply understanding of a personal name is crucial for personal name disambiguation.

Language may be viewed as a system of conventionalized cognition in the sense that it conventionally conceptualizes a range of perceived realities shared by a community of speakers. In this talk, I focus on how language, and Chinese in particular, conventionalizes cognition in order to contribute to a macro-theory of cognitive informatics. Based on distributional data from a large-scale corpus, I will show that Chinese conventionally and selectively represents important aspects of cognition, such as transition vs. state, and production vs. perception. I will also draw attention to the fact that these conventions lead to the packaging of cognitive events, i.e., following Pustejovsky's event coercion, which can be seen as an extension of Aristotle's four causes of knowledge. Lastly, relying on this intuition that language conventionalizes cognition, I will propose a simple yet robust approach to text-based emotion detection and classification.

We offer a preliminary study of how knowledge may be represented differently in different languages. In particular, we account for the contrast between English and Chinese when identical target domain knowledge is represented with two different, yet related, source domains in each language. We incorporate corpora analysis in English and Chinese with SUMO to delimit the source domains identified using the conceptual mapping model (Ahrens. 2002). In particular, we investigate economy metaphors with the source domains of aeroplane and moving vehicle. These two source domains are found in Chinese and English respectively. Hence, we ask the question whether these source domains should be conflated under a general source domain such as TransportationDevice or should they form independent source domains. Our study addresses this issue by using the SUMO ontology as well as the mapping principles found in the corpora analysis. Our studies contribute to automatizing the source-target domain mappings in conceptual metaphors.

The sinica BOW (academia sinica bilingual ontological WordNet) is intended as a linguistic infrastructure for knowledge representation and knowledge engineering. It is built upon the relation-based structure of WordNet. On one hand, a bilingual WordNet is constructed with the crucial design feature of treating bilingual translation correspondences as lexical semantic relations. On the other hand, SUMO (suggested upper merged ontology) is adopted as the shared system of conceptual categorization. SUMO is also one of the first conceptual categorization systems to be mapped to an English lexicon. By the combination of ontology and WordNet, we hope that sinica BOW will 1) give each linguistic form a rigorous conceptual location, 2) clarify the relation between conceptual classification and linguistic instantiation, and 3) facilitate genuine cross-lingual access of knowledge. In addition to the integration of WordNet and ontology, it is also an important goal of sinica BOW to integrate lexical resources such that the linguistic-to-conceptual representation and language -to- language gaps can be bridged simultaneously. Taking lexicon as a bridging knowledgebase and ontology as the overall knowledge structure seems to be a logical choice. Integrating the two resources with multilingual capacity will add to the versatility and open new possibilities.