Explore the vast range of titles available.

Hate Speech in social media is a complex phenomenon, whose detection has recently gained significant traction in the Natural Language Processing community, as attested by several recent review works. Annotated corpora and benchmarks are key resources, considering the vast number of supervised approaches that have been proposed. Lexica play an important role as well for the development of hate speech detection systems. In this review, we systematically analyze the resources made available by the community at large, including their development methodology, topical focus, language coverage, and other factors. The results of our analysis highlight a heterogeneous, growing landscape, marked by several issues and venues for improvement.

Recently, artificial intelligence (AI)-programmed automated writing evaluation (AWE) has attracted increasing attention in language research. Using a small data set arising from an analysis of five Chinese university-level English as a foreign language (EFL) students’ submissions, this paper examined in detail how EFL students interacted with the feedback of Pigai, the largest AI-programmed AWE in China. The analysis started with the intention of capturing the machine feedback on the five students’ submissions and the exchanges between the participants and Pigai over repeated submissions, ranging from 3 to 12 submissions. The analysis showed that the learners’ interactions with Pigai focused on error corrective feedback in the initial two submissions. In the case of one student who had 12 submissions, the non-error corrective feedback increased gradually over time, providing rich linguistic resources but without examples and contextual information. The students’ take-up rates of feedback with linguistic resources were much lower than that of error corrective and general feedback. A terrain model to map the stages and nature of student responses showed a more complete dynamic process, in which students’ responses changed from the initial mechanical responses at the discrete language level to more considered approaches in response to machine feedback. The findings of this study have implications for both language pedagogy and the future design and development of AWE for second or foreign language learning.

Irony is a pervasive aspect of many online texts, one made all the more difficult by the absence of face-to-face contact and vocal intonation. As our media increasingly become more social, the problem of irony detection will become even more pressing. We describe here a set of textual features for recognizing irony at a linguistic level, especially in short texts created via social media such as Twitter postings or \"tweets\". Our experiments concern four freely available data sets that were retrieved from Twitter using content words (e.g. \"Toyota\") and user-generated tags (e.g. \"#irony\"). We construct a new model of irony detection that is assessed along two dimensions: representativeness and relevance. Initial results are largely positive, and provide valuable insights into the figurative issues facing tasks such as sentiment analysis, assessment of online reputations, or decision making.

In this article, we describe a method for automatically detecting syllable nuclei in order to measure speech rate without the need for a transcription. A script written in the software program Praat (Boersma & Weenink, 2007) detects syllables in running speech. Peaks in intensity (dB) that are preceded and followed by dips in intensity are considered to be potential syllable nuclei. The script subsequently discards peaks that are not voiced. Testing the resulting syllable counts of this script on two corpora of spoken Dutch, we obtained high correlations between speech rate calculated from human syllable counts and speech rate calculated from automatically determined syllable counts. We conclude that a syllable count measured in this automatic fashion suffices to reliably assess and compare speech rates between participants and tasks.

The detection of online cyberbullying has seen an increase in societal importance, popularity in research, and available open data. Nevertheless, while computational power and affordability of resources continue to increase, the access restrictions on high-quality data limit the applicability of state-of-the-art techniques. Consequently, much of the recent research uses small, heterogeneous datasets, without a thorough evaluation of applicability. In this paper, we further illustrate these issues, as we (i) evaluate many publicly available resources for this task and demonstrate difficulties with data collection. These predominantly yield small datasets that fail to capture the required complex social dynamics and impede direct comparison of progress. We (ii) conduct an extensive set of experiments that indicate a general lack of cross-domain generalization of classifiers trained on these sources, and openly provide this framework to replicate and extend our evaluation criteria. Finally, we (iii) present an effective crowdsourcing method: simulating real-life bullying scenarios in a lab setting generates plausible data that can be effectively used to enrich real data. This largely circumvents the restrictions on data that can be collected, and increases classifier performance. We believe these contributions can aid in improving the empirical practices of future research in the field.

The Leximancer system is a relatively new method for transforming lexical co-occurrence information from natural language into semantic patterns in a nunsupervised manner. It employs two stages of co-occurrence information extraction-semantic and relational-using a different algorithm for each stage. The algorithms used are statistical, but they employ nonlinear dynamics and machine learning. This article is an attempt to validate the output of Leximancer, using a set of evaluation criteria taken from content analysis that are appropriate for knowledge discovery tasks.

Purpose: The evaluation of spelling performance in aphasia reveals deficits in written language and can facilitate the design of targeted writing treatments. Nevertheless, manual scoring of spelling performance is time-consuming, laborious, and error prone. We propose a novel method based on the use of distance metrics to automatically score spelling. This study compares six automatic distance metrics to identify the metric that best corresponds to the gold standard--manual scoring--using data from manually obtained spelling scores from individuals with primary progressive aphasia. Method: Three thousand five hundred forty word and nonword spelling productions from 42 individuals with primary progressive aphasia were scored manually. The gold standard--the manual scores--were compared to scores from six automated distance metrics: sequence matcher ratio, Damerau-Levenshtein distance, normalized Damerau-Levenshtein distance, Jaccard distance, Masi distance, and Jaro-Winkler similarity distance. We evaluated each distance metric based on its correlation with the manual spelling score. Results: All automatic distance scores had high correlation with the manual method for both words and nonwords. The normalized Damerau-Levenshtein distance provided the highest correlation with the manual scoring for both words (r[subscript s] = 0.99) and nonwords (r[subscript s] = 0.95). Conclusions: The high correlation between the automated and manual methods suggests that automatic spelling scoring constitutes a quick and objective approach that can reliably substitute the existing manual and time-consuming spelling scoring process, an important asset for both researchers and clinicians.

With the increasing popularity of user-generated content on social media, the number of toxic texts is also on the rise. Such texts cause adverse effects on users and society at large, therefore, the identification of toxic comments is a growing need of the day. While toxic comment classification has been studied for resource-rich languages like English, no work has been done for Roman Urdu despite being a widely used language on social media in South Asia. This paper addresses the challenge of Roman Urdu toxic comment detection by developing a first-ever large labeled corpus of toxic and non-toxic comments. The developed corpus, called RUT (Roman Urdu Toxic), contains over 72 thousand comments collected from popular social media platforms and has been labeled manually with a strong inter-annotator agreement. With this dataset, we train several classification models to detect Roman Urdu toxic comments, including classical machine learning models with the bag-of-words representation and some recent deep models based on word embeddings. Despite the success of the latter in classifying toxic comments in English, the absence of pre-trained word embeddings for Roman Urdu prompted to generate different word embeddings using Glove, Word2Vec and FastText techniques, and compare them with task-specific word embeddings learned inside the classification task. Finally, we propose an ensemble approach, reaching our best F1-score of 86.35%, setting the first-ever benchmark for toxic comment classification in Roman Urdu.

Swearing plays an ubiquitous role in everyday conversations among humans, both in oral and textual communication, and occurs frequently in social media texts, typically featured by informal language and spontaneous writing. Such occurrences can be linked to an abusive context, when they contribute to the expression of hatred and to the abusive effect, causing harm and offense. However, swearing is multifaceted and is often used in casual contexts, also with positive social functions. In this study, we explore the phenomenon of swearing in Twitter conversations, by automatically predicting the abusiveness of a swear word in a tweet as the main investigation perspective. We developed the Twitter English corpus SWAD (Swear Words Abusiveness Dataset), where abusive swearing is manually annotated at the word level. Our collection consists of 2577 instances in total from two phases of manual annotation. We developed models to automatically predict abusive swearing, to provide an intrinsic evaluation of SWAD and confirm the robustness of the resource. We model this prediction task as three different tasks, namely sequence labeling, text classification, and target-based swear word abusiveness prediction. We experimentally found that our intention to model the task similarly to aspect-based sentiment analysis leads to promising results. Subsequently, we employ the classifier to improve the prediction of abusive language in several standard benchmarks. The results of our experiments show that additional abusiveness feature of the swear words is able to improve the performance of abusive language detection models in several benchmark datasets.

Abusive language is becoming a problematic issue for our society. The spread of messages that reinforce social and cultural intolerance could have dangerous effects in victims’ life. State-of-the-art technologies are often effective on detecting explicit forms of abuse, leaving unidentified the utterances with very weak offensive language but a strong hurtful effect. Scholars have advanced theoretical and qualitative observations on specific indirect forms of abusive language that make it hard to be recognized automatically. In this work, we propose a battery of statistical and computational analyses able to support these considerations, with a focus on creative and cognitive aspects of the implicitness, in texts coming from different sources such as social media and news. We experiment with transformers, multi-task learning technique, and a set of linguistic features to reveal the elements involved in the implicit and explicit manifestations of abuses, providing a solid basis for computational applications.