Search

In the following paper, we investigated the usefulness of future reference sentence patterns in the prediction of the unfolding of future events. To obtain such patterns we first collected sentences that have any reference to the future from newspapers and Web news. Based on this collection, we developed a novel method for automatic extraction of frequent patterns from such sentences. The extracted patterns, consisting of multilayer semantic information and morphological information, were implemented in the formation of a general model of linguistically expressed future. To fully assess the performance of the proposed method we performed a number of evaluation experiments. In the first experiment, we evaluated the automatic extraction of future reference sentence patterns with the proposed extraction algorithm. In the second set of experiments, we estimated the effectiveness of those patterns and applied them to automatically classify sentences into future referring and other. The final model was then tested for performance in retrieving a new set of future reference sentences from a large news corpus. The obtained results confirmed that the proposed method outperformed state-of-the-art method in fully automatic retrieval of future reference sentences. Lastly, we applied the method in practice to confirm its usefulness in two tasks. The first is to support human readers in the everyday prediction of unfolding future events. In the second task, we developed a fully automatic prototype method for future prediction and tested its performance using the tasks included in the official Future Prediction Competence Test. The results indicate that the prototype system outperforms natural human foreseeing capability.

In this paper, we study language used by suicidal users on Reddit social media platform. To do that, we firstly collect a large-scale dataset of Reddit posts and annotate it with highly trained and expert annotators under a rigorous annotation scheme. Next, we perform a multifaceted analysis of the dataset, including: (1) the analysis of user activity before and after posting a suicidal message, and (2) a pragmalinguistic study on the vocabulary used by suicidal users. In the second part of the analysis, we apply LIWC, a dictionary-based toolset widely used in psychology and linguistic research, which provides a wide range of linguistic category annotations on text. However, since raw LIWC scores are not sufficiently reliable, or informative, we propose a procedure to decrease the possibility of unreliable and misleading LIWC scores leading to misleading conclusions by analyzing not each category separately, but in pairs with other categories. The analysis of the results supported the validity of the proposed approach by revealing a number of valuable information on the vocabulary used by suicidal users and helped to pin-point false predictors. For example, we were able to specify that death-related words, typically associated with suicidal posts in the majority of the literature, become false predictors, when they co-occur with apostrophes, even in high-risk subreddits. On the other hand, the category-pair based disambiguation helped to specify that death becomes a predictor only when co-occurring with future-focused language, informal language, discrepancy, or 1st person pronouns. The promising applicability of the approach was additionally analyzed for its limitations, where we found out that although LIWC is a useful and easily applicable tool, the lack of any contextual processing makes it unsuitable for application in psychological and linguistic studies. We conclude that disadvantages of LIWC can be easily overcome by creating a number of high-performance AI-based classifiers trained for annotation of similar categories as LIWC, which we plan to pursue in future work.

In this paper, we present a Deep Learning-based system for the support of information triaging on Twitter during emergency situations, such as disasters, or other influential events, such as political elections. The system is based on the assumption that a different type of information is required right after the event and some time after the event occurs. In a preliminary study, we analyze the language behavior of Twitter users during two kinds of influential events, namely, natural disasters and political elections. In the study, we analyze the credibility of information included by users in tweets in the above-mentioned situations, by classifying the information into two kinds: Primary Information (first-hand reports) and Secondary Information (second-hand reports, retweets, etc.). We also perform sentiment analysis of the data to check user attitudes toward the occurring events. Next, we present the structure of the system and compare a number of classifiers, including the proposed one based on Convolutional Neural Networks. Finally, we validate the system by performing an in-depth analysis of information obtained after a number of additional events, including an eruption of a Japanese volcano Ontake on 27 September 2014, as well as heavy rains and typhoons that occurred in 2020. We confirm that the methods works sufficiently well even when trained on data from nearly 10 years ago, which strongly suggests that the model is well-generalized and sufficiently grasps important aspects of each type of classified information.

Cyberbullying, or humiliating people using the Internet, has existed almost since the beginning of Internet communication.The relatively recent introduction of smartphones and tablet computers has caused cyberbullying to evolve into a serious social problem. In Japan, members of a parent-teacher association (PTA)attempted to address the problem by scanning the Internet for cyber bullying entries. To help these PTA members and other interested parties confront this difficult task we propose a novel method for automatic detection of malicious Internet content. This method is based on a combinatorial approach resembling brute-force search algorithms, but applied in language classification. The method extracts sophisticated patterns from sentences and uses them in classification. The experiments performed on actual cyberbullying data reveal an advantage of our method vis-à-visprevious methods. Next, we implemented the method into an application for Android smartphones to automatically detect possible harmful content in messages. The method performed well in the Android environment, but still needs to be optimized for time efficiency in order to be used in practice

In this paper we present our research devoted to the development of Natural Language Processing technologies for the Ainu language, a critically endangered language isolate spoken by the Ainu people, the native inhabitants of northern parts of the Japanese archipelago. In particular, we focused on improving the existing tools for transcription normalization, word segmentation (tokenization) and part-of-speech tagging. In the experiments we applied two Ainu language dictionaries from different domains (literary and colloquial) and created a new data set by combining them. The experiments confirmed the positive effect of these modifications on the overall performance of the tools, especially with objective samples unrelated to the training data. We also discuss further improvements obtained by applying corpus-driven language models to the problem of word segmentation and using a state-of-the-art tool for training part-of-speech taggers.

We describe our attempt to apply a state-of-the-art sequential tagger – SVMTool – in the task of automatic part-of-speech annotation of the Ainu language, a critically endangered language isolate spoken by the native inhabitants of northern Japan. Our experiments indicated that it performs better than the custom system proposed in previous research (POST-AL), especially when applied to out-of-domain data. The biggest advantage of the model trained using SVMTool over the POST-AL tagger is its ability to guess part-of-speech tags for OoV words, with the accuracy of up to 63%.

In this paper we introduce HEMOS (Humor-EMOji-Slang-based) system for fine-grained sentiment classification for the Chinese language using deep learning approach. We investigate the importance of recognizing the influence of humor, pictograms and slang on the task of affective processing of the social media. In the first step, we collected 576 frequent Internet slang expressions as a slang lexicon; then, we converted 109 Weibo emojis into textual features creating a Chinese emoji lexicon. In the next step, by performing two polarity annotations with new “optimistic humorous type” and “pessimistic humorous type” added to standard “positive” and “negative” sentiment categories, we applied both lexicons to attention-based bi-directional long short-term memory recurrent neural network (AttBiLSTM) and tested its performance on undersized labeled data. Our experimental results show that the proposed method can significantly improve the state-of-the-art methods in predicting sentiment polarity on Weibo, the largest Chinese social network.

Pictograms (emoticons/emojis) have been widely used in social media as a mean for graphical expression of emotions. People can express delicate nuances through textual information when supported with emoticons, and the effectiveness of computer-mediated communication (CMC) is also improved. Therefore it is important to fully understand the influence of emoticons on CMC. In this paper, we propose an emoticon polarity-aware recurrent neural network method for sentiment analysis of Weibo, a Chinese social media platform. In the first step, we analyzed the usage of 67 emoticons with racial expression used on Weibo. By performing a polarity annotation with a new “humorous type” added, we have confirmed that 23 emoticons can be considered more as humorous than positive or negative. On this basis, we applied the emoticons polarity in a Long Short-Term Memory recurrent neural network (LSTM) for sentiment analysis of undersized labelled data. Our experimental results show that the proposed method can significantly improve the precision for predicting sentiment polarity on Weibo.

This paper presents our research on automatic annotation of a five-billion-word corpus of Japanese blogs with information on affect and sentiment. We first perform a study in emotion blog corpora to discover that there has been no large scale emotion corpus available for the Japanese language. We choose the largest blog corpus for the language and annotate it with the use of two systems for affect analysis: ML-Ask for word- and sentence-level affect analysis and CAO for detailed analysis of emoticons. The annotated information includes affective features like sentence subjectivity (emotive/non-emotive) or emotion classes (joy, sadness, etc.), useful in affect analysis. The annotations are also generalized on a two-dimensional model of affect to obtain information on sentence valence (positive/negative), useful in sentiment analysis. The annotations are evaluated in several ways. Firstly, on a test set of a thousand sentences extracted randomly and evaluated by over forty respondents. Secondly, the statistics of annotations are compared to other existing emotion blog corpora. Finally, the corpus is applied in several tasks, such as generation of emotion object ontology or retrieval of emotional and moral consequences of actions.

This paper presents our research in detection of emotive (emotionally loaded) sentences. The task is defined as a text classification problem with an assumption that emotive sentences stand out both lexically and grammatically. The assumption is verified exper- imentally. The experiment is based on n-grams as well as more sophisticated patterns with disjointed elements. To deal with the sophisticated patterns a novel language modelling algorithm based on the idea of language combinatorics is applied. The results of experiments are explained with the standard means of Precision, Recall and balanced F-score. The algorithm also provides a refined list of most frequent sophisticated patterns typical for both emotive and non-emotive context.

Ainu is a critically endangered language spoken by the native inhabitants of northern Japan. This paper describes our research aimed at the development of technology for automatic processing of text in Ainu. In particular, we improved the existing tools for normalizing old transcriptions, word segmentation, and part-of-speech tagging. In the experiments we applied two Ainu language dictionaries from different domains (literary and colloquial) and created a new data set by combining them. The experiments revealed that expanding the lexicon had a positive impact on the overall performance of our tools, especially with test data unrelated to any of the training sets used.

Word segmentation is an essential task in automatic language processing for languages where there are no explicit word boundary markers, or where space-delimited orthographic words are too coarse-grained. In this paper we introduce the MiNgMatch Segmenter&mdash, a fast word segmentation algorithm, which reduces the problem of identifying word boundaries to finding the shortest sequence of lexical n-grams matching the input text. In order to validate our method in a low-resource scenario involving extremely sparse data, we tested it with a small corpus of text in the critically endangered language of the Ainu people living in northern parts of Japan. Furthermore, we performed a series of experiments comparing our algorithm with systems utilizing state-of-the-art lexical n-gram-based language modelling techniques (namely, Stupid Backoff model and a model with modified Kneser-Ney smoothing), as well as a neural model performing word segmentation as character sequence labelling. The experimental results we obtained demonstrate the high performance of our algorithm, comparable with the other best-performing models. Given its low computational cost and competitive results, we believe that the proposed approach could be extended to other languages, and possibly also to other Natural Language Processing tasks, such as speech recognition.

This paper presents our research in text-based affect analysis (AA) of narratives. AA represents a task of estimating or recognizing emotions elicited by a certain semiotic modality. In text-based AA the modality in focus is the textual representation of language. In this research we study particularly one type of language realization, namely narratives (e.g., stories, fairy tales, etc.). Affect analysis within the context of narratives is a challenging task because narratives are created of different kinds of sentences (descriptions, dialogs, etc.). Moreover, different characters become subjects of different emotional expressions in different parts of narratives. In this research we address the problem of person/character related affect recognition in narratives. We propose a method for emotion subject extraction from a sentence based on analysis of anaphoric expressions and compare two methods for affect analysis. We evaluate the system and discuss its possible future improvements.

This paper presents POST-AL, the first part-of-speech tagger for Ainu language. The system uses a hand-crafted dictionary based on Ainu narratives ''yukar''. The system provides three types of information: word/token, part of speech, and translation of the token (in Japanese). Evaluation on a training set provided positive results. The system could be useful in a great number of tasks related to the research on Ainu language, such as content analysis or translation, which till now have been done mostly manually.

This paper presents CAO, a system for affect analysis of emoticons in Japanese online communication. Emoticons are strings of symbols widely used in text-based online communication to convey user emotions. The presented system extracts emoticons from input and determines the specific emotion types they express with a three-step procedure. First, it matches the extracted emoticons to a predetermined raw emoticon database. The database contains over 10,000 emoticon samples extracted from the Web and annotated automatically. The emoticons for which emotion types could not be determined using only this database, are automatically divided into semantic areas representing “mouths” or “eyes,” based on the idea of kinemes from the theory of kinesics. The areas are automatically annotated according to their co-occurrence in the database. The annotation is first based on the eye-mouth-eye triplet, and if no such triplet is found, all semantic areas are estimated separately. This provides hints about potential groups of expressed emotions, giving the system coverage exceeding 3 million possibilities. The evaluation, performed on both training and test sets, confirmed the system's capability to sufficiently detect and extract any emoticon, analyze its semantic structure, and estimate the potential emotion types expressed. The system achieved nearly ideal scores, outperforming existing emoticon analysis systems.

We propose a method for affect analysis of textual input in Japanese supported with Web mining. The method is based on a pragmatic reasoning that emotional states of a speaker are conveyed by emotional expressions used in emotive utterances. It means that if an emotive expression is used in a sentence in a context described as emotive, the emotion conveyed in the text is revealed by the used emotive expression. The system ML-Ask (Emotive Elements / Expressions Analysis System) is constructed on the basis of this idea. An evaluation of the system is performed in which two evaluation methods are compared. To choose the most objective evaluation method we compare the most popular method in the field and a method proposed by us. The proposed evaluation method was shown to be more objective and revealed the strong and weak points of the system in detail. In the evaluation experiment ML-Ask reached human level in recognizing the general emotiveness of an utterance (0.83 balanced F-score) and 63% of human level in recognizing the specific types of emotions. We support the system with a Web mining technique to improve the performance of emotional state types extraction. In the Web mining technique emotive associations are extracted from the Web using co-occurrences of emotive expressions with morphemes of causality. The Web mining technique improved the performance of the emotional states types extraction to 85% of human performance.

In this research we focus on discriminating between emotive (emotionally loaded) and non-emotive sentences. We define the problem from a linguistic point of view assuming that emotive sentences stand out both lexically and grammatically. We verify this assumption experimentally by comparing two sets of such sentences in Japanese. The comparison is based on words, longer n-grams as well as more sophisticated patterns. In the classification weuse a novelunsupervised learning algorithm based on the idea of language combinatorics. The method reached results comparable to the state of the art, while the fact that it is fully automatic makes it more efficient and language independent.

Emoticons are string of symbols representing body language in text-based communication. For a long time they have been considered as unnatural language entities. This chapter argues that, in over 40-year-long history of text-based communication, emoticons have gained a status of an indispensable means of support for text-based messages. This makes them fully a part of Natural Language Processing. The fact the emoticons have been considered as unnatural language expressions has two causes. Firstly, emoticons represent body language, which by definition is nonverbal. Secondly, there has been a lack of sufficient methods for the analysis of emoticons. Emoticons represent a multimodal (bimodal in particular) type of information. Although they are embedded in lexical form, they convey non-linguistic information. To prove this argument the authors propose that the analysis of emoticons was based on a theory designed for the analysis of body language. In particular, the authors apply the theory of kinesics to develop a state of the art system for extraction and analysis of kaomoji, Japanese emoticons. The system performance is verified in comparison with other emoticon analysis systems. Experiments showed that the presented approach provides nearly ideal results in different aspects of emoticon analysis, thus proving that emoticons possess features of multimodal expressions.

Humor processing is still a less studied issue, both in NLP and AI. In this paper we contribute to this field. In our previous research we showed that adding a simple pun generator to a chatterbot can significantly improve its performance. The pun generator we used generated only puns based on words (not phrases). In this paper we introduce the next stage of the system's development—an algorithm allowing generation of phrasal pun candidates. We show that by using only the Internet (without any hand-made humor-oriented lexicons), it is possible to generate puns based on complex phrases. As the output list is often excessively long, we also propose a method for reducing the number of candidates by comparing two web-query-based rankings. The evaluation experiment showed that the system achieved an accuracy of 72.5% for finding proper candidates in general, and the reduction method allowed us to significantly shorten the candidates list. The parameters of the reduction algorithm are variable, so that the balance between the number of candidates and the quality of output can be manipulated according to needs.

By our demonstration we want to introduce our achievements in combining different purpose algorithms to build a chatbot which is able to keep a conversation on any topic. It uses snippets of Internet search results to stay within a context, Nakamura's Emotion Dictionary to detect an emotional load existence and categorization of a textual utterance and a causal consequences retrieval algorithm when emotive features are not found. It is also able to detect a possibility to make a pun by analyzing the input sentence and create one if timing is adequate.

This paper contains the results of evaluation experiments conducted to investigate if implementation of a pun generator into a non-task oriented talking system improves the latter's performance. We constructed a simple joking conversational system and conducted one user evaluation experiment and two third person evaluation experiments. The results showed that humor does have a positive influence on the dialogue between humans and computers. The implications of this fact and problems that occurred during the research are discussed. We also propose how they can be solved in the future.

In this paper we propose a novel method for automatic evaluation of conversational agents. The method is based on analyzing the userpsilas affect conveyed in utterances. From analyzing: the userpsilas general emotional engagement in the conversation and the emotion types conveyed by the user in the conversation, a simple psychological reasoning is derived about the userpsilas sentiment about the agentpsilas performance. The evaluation experiment on two Japanese-speaking conversational agents showed the same tendencies in the results returned by the system constructed on the proposed method and the userpsilas opinion about the two agents checked in the afterward survey. Thus the method can be used for evaluation of Japanese-speaking conversational agents.

This paper presents a novel method for estimating speaker's affective states based on two contextual features: valence shifters and appropriateness. Firstly, a system for affect analysis is used to recognise specific types of emotions. We improve the baseline system with the analysis of Contextual Valence Shifters (CVS), which determine the semantic orientation of emotive expressions. Secondly, a web mining technique is used to verify the appropriateness of the recognised emotions for the particular context. Verification of contextual appropriateness of emotions is the next step towards implementation of Emotional Intelligence Framework in machines. The proposed method is evaluated using two conversational agents.