Search

In this paper, we investigate sarcasm and irony as seen through a novel perspective of machine translation. We employ various techniques for translation, comparing both manually and automatically translated datasets of irony and sarcasm. We first clarify the definitions of irony and sarcasm and present an exhaustive field review of studies on irony both from purely linguistic as well as computational linguistic perspectives. We also propose a novel evaluation metric for the purpose of evaluating translations of figurative language, with a focus on machine-translated irony and sarcasm. The constructed English and Chinese parallel dataset includes polarized content from tweets as well as forum posts, categorized by irony types. The preferred translation model, mBART-50, is identified through a thorough experimental process. Optimal translation settings and the best-finetuned model for irony are explored, with the most effective model being finetuned on both ironic and non-ironic data. We also experimented which types of irony are best suitable for training in this specific task — short microblogging messages or longer forum posts. Moreover, we compare the capabilities of a well fine-tuned mBART to a prompt-based method using the recently popular ChatGPT model, with the conclusion that the former still outperforms the latter, although ChatGPT without any training can be considered as a “good enough” ad hoc solution in the case of a lack of data for training. Finally, we verify if the translated data – either manually, or with an MT model – can be used as training data in a task of irony detection. We believe that the presented research can be expanded into languages other than the presented here Chinese and English, which together with the ability to detect various categories of irony, could contribute to deepening the understanding of figurative language, especially irony and sarcasm.

This study explores the feasibility of using artificial emotional speech datasets generated by existing artificial voice-generating software as an alternative to human-generated datasets for emotional speech synthesis. Focusing on the Japanese language, we assess the viability of these artificial datasets in languages with limited emotional speech resources. Our approach combines qualitative and quantitative analyses to evaluate the effectiveness of synthetic emotional speech in replicating human-like emotional expression. The results demonstrate that while artificial datasets can approximate certain emotional states, there are significant limitations in replicating the full range of human emotions, particularly in subtle or mixed emotions. These findings underscore the potential and current constraints of using artificial datasets in emotional speech synthesis, suggesting avenues for future research to enhance the quality and emotional expressiveness of synthetic speech.

False news refers to false, fake, or misleading information presented as real news. In recent years, there has been a noticeable increase in false news on the Internet. The goal of this paper was to study the automatic detection of such false news using machine learning and natural language processing techniques and to determine which techniques work the most effectively. This article first studies what constitutes false news and how it differs from other types of misleading information. We also study the results achieved by other researchers on the same topic. After building a foundation to understand false news and the various ways of automatically detecting it, this article provides its own experiments. These experiments were carried out on four different datasets, one that was made just for this article, using 10 different machine learning methods. The results of this article were satisfactory and provided answers to the original research questions set up at the beginning of this article. This article could determine from the experiments that passive aggressive algorithms, support vector machines, and random forests are the most efficient methods for automatic false news detection. This article also concluded that more complex experiments, such as using multiple levels of identifying false news or detecting computer-generated false news, require more complex machine learning models.

The rapid advancement of artificial intelligence (AI) and natural language processing (NLP) has profoundly impacted our understanding of emotions, decision-making, and opinions, particularly within the context of the Internet and social media [...]

Wide adoption of social media has caused an explosion of information stored online, with the majority of that information containing subjective, opinionated, and emotional content produced daily by users. The field of emotion analysis has helped effectively process such human emotional expressions expressed in daily social media posts. Unfortunately, one of the greatest limitations of popular word-based emotion analysis systems has been the limited emotion vocabulary. This paper presents an attempt to extensively expand one such word-based emotion analysis system by integrating multiple emotion dictionaries and implementing an automatic extraction mechanism for emotive expressions. We first leverage diverse emotive expression dictionaries to expand the emotion lexicon of the system. To do that, we solve numerous problems with the integration of various dictionaries collected using different standards. We demonstrate the performance improvement of the system with improved accuracy and granularity of emotion classification. Furthermore, our automatic extraction mechanism facilitates the identification of novel emotive expressions in an emotion dataset, thereby enriching the depth and breadth of emotion analysis capabilities. In particular, the automatic extraction method shows promising results for applicability in further expansion of the dictionary base in the future, thus advancing the field of emotion analysis and offering new avenues for research in sentiment analysis, affective computing, and human–computer interaction.

We introduce the first dataset of harmful and offensive language collected from the Polish Internet. This dataset was meticulously curated to facilitate the exploration of harmful online phenomena such as cyberbullying and hate speech, which have exhibited a significant surge both within the Polish Internet as well as globally. The dataset was systematically collected and then annotated using two approaches. First, it was annotated by two proficient layperson volunteers, operating under the guidance of a specialist in the language of cyberbullying and hate speech. To enhance the precision of the annotations, a secondary round of annotations was carried out by a team of adept annotators with specialized long-term expertise in cyberbullying and hate speech annotations. This second phase was further overseen by an experienced annotator, acting as a super-annotator. In its initial application, the dataset was leveraged for the categorization of cyberbullying instances in the Polish language. Specifically, the dataset serves as the foundation for two distinct tasks: (1) a binary classification that segregates harmful and non-harmful messages and (2) a multi-class classification that distinguishes between two variations of harmful content (cyberbullying and hate speech), as well as a non-harmful category. Alongside the dataset itself, we also provide the models that showed satisfying classification performance. These models are made accessible for third-party use in constructing cyberbullying prevention systems.

The proliferation of the internet, especially on social media platforms, has amplified the prevalence of cyberbullying and harassment. Addressing this issue involves harnessing natural language processing (NLP) and machine learning (ML) techniques for the automatic detection of harmful content. However, these methods encounter challenges when applied to low-resource languages like the Chittagonian dialect of Bangla. This study compares two approaches for identifying offensive language containing vulgar remarks in Chittagonian. The first relies on basic keyword matching, while the second employs machine learning and deep learning techniques. The keyword-matching approach involves scanning the text for vulgar words using a predefined lexicon. Despite its simplicity, this method establishes a strong foundation for more sophisticated ML and deep learning approaches. An issue with this approach is the need for constant updates to the lexicon. To address this, we propose an automatic method for extracting vulgar words from linguistic data, achieving near-human performance and ensuring adaptability to evolving vulgar language. Insights from the keyword-matching method inform the optimization of machine learning and deep learning-based techniques. These methods initially train models to identify vulgar context using patterns and linguistic features from labeled datasets. Our dataset, comprising social media posts, comments, and forum discussions from Facebook, is thoroughly detailed for future reference in similar studies. The results indicate that while keyword matching provides reasonable results, it struggles to capture nuanced variations and phrases in specific vulgar contexts, rendering it less robust for practical use. This contradicts the assumption that vulgarity solely relies on specific vulgar words. In contrast, methods based on deep learning and machine learning excel in identifying deeper linguistic patterns. Comparing SimpleRNN models using Word2Vec and fastText embeddings, which achieved accuracies ranging from 0.84 to 0.90, logistic regression (LR) demonstrated remarkable accuracy at 0.91. This highlights a common issue with neural network-based algorithms, namely, that they typically require larger datasets for adequate generalization and competitive performance compared to conventional approaches like LR.

When planning a travel or an adventure, sightseers increasingly rely on opinions posted on the Internet tourism related websites, such as TripAdvisor, Booking.com or Expedia. Unfortunately, beautiful, yet less-known places and rarely visited sightspots often do not accumulate sufficient number of valuable opinions on such websites. On the other hand, users often post their opinions on casual social media services, such as Facebook, Instagram or Twitter. Therefore, in this study, we develop a system for supplementing insufficient number of Internet opinions available for sightspots with tweets containing opinions of such sightspots, with a specific focus on wildlife sightspots. To do that, we develop an approach consisting of a system (PSRS) for wildlife sightspots and propose a method for verifying collected geotagged tweets and using them as on-spot reviews. Tweets that contain geolocation information are considered geotagged and therefore treated as possible tourist on-spot reviews. The main challenge, however, is to confirm the authenticity of the extracted tweets. Our method includes the use of location clustering and classification techniques. Specifically, extracted geotagged tweets are clustered by using location information and then annotated taking into consideration specific features applied to machine learning-based classification techniques. As for the machine learning (ML) algorithms, we adopt a fine-tuned transformer neural network-based BERT model which implements the information of token context orientation. The BERT model achieved a higher F-score of 0.936, suggesting that applying a state-of-the-art deep learning-based approach had a significant impact on solving this task. The extracted tweets and annotated scores are then mapped on the designed Park Supplementary Review System (PSRS) as supplementary reviews for travelers seeking additional information about the related sightseeing spots.

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.

We present ML-Ask – the first Open Source Affect Analysis system for textual input in Japanese. ML-Ask analyses the contents of an input (e.g., a sentence) and annotates it with information regarding the contained general emotive expressions, specific emotional words, valence-activation dimensions of overall expressed affect, and particular emotion types expressed with their respective expressions. ML-Ask also incorporates the Contextual Valence Shifters model for handling negation in sentences to deal with grammatically expressible shifts in the conveyed valence. The system, designed to work mainly under Linux and MacOS, can be used for research on, or applying the techniques of Affect Analysis within the framework Japanese language. It can also be used as an experimental baseline for specific research in Affect Analysis, and as a practical tool for written contents annotation. Funding statement: This research has been supported by: a Research Grant from the Nissan Science Foundation (years 2009–2010), The GCOE Program founded by Japan’s Ministry of Education, Culture, Sports, Science and Technology (years 2009–2010), (JSPS) KAKENHI Grant-in-Aid for JSPS Fellows (Project Number: 22-00358) (years 2010–2012), (JSPS) KAKENHI Grant-in-Aid for Scientific Research (Project Number: 24600001) (years 2012–2015), (JSPS) KAKENHI Grant-in-Aid for Research Activity Start-up (Project Number: 25880003) (years 2013–2015), and (JSPS) KAKENHI Grant-in-Aid for Encouragement of Young Scientists (B) (Project Number: 15K16044) (years 2015-present, project estimated to end in March 2018).

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—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.

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.