Investigating cross-lingual training for offensive language detection.
Ontology highlight
ABSTRACT:
SUBMITTER: Pelicon A
PROVIDER: S-EPMC8237322 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Social media platforms have become inundated with offensive language. This issue must be addressed for the growth of online social networks (OSNs) and a healthy online environment. While significant research has been devoted to identifying toxic content in major languages like English, this remains an open area of research in the low-resource Pashto language. This study aims to develop an AI model for the automatic detection of offensive textual content in Pashto. To achieve this goal, we have developed a benchmark dataset called the Pashto Offensive Language Dataset (POLD), which comprises tweets collected from Twitter and manually classified into two categories: "offensive" and "not offensive". To discriminate these two categories, we investigated the classic deep learning classifiers based on neural networks, including CNNs and RNNs, using static word embeddings: Word2Vec, fastText, and GloVe as features. Furthermore, we examined two transfer learning approaches. In the first approach, we fine-tuned the pre-trained multilingual language model, XLM-R, using the POLD dataset, whereas, in the second approach, we trained a monolingual BERT model for Pashto from scratch using a custom-developed text corpus. Pashto BERT was then fine-tuned similarly to XLM-R. The performance of all the deep learning and transformer learning models was evaluated using the POLD dataset. The experimental results demonstrate that our pre-trained Pashto BERT model outperforms the other models, achieving an F1-score of 94.34% and an accuracy of 94.77%.
| S-EPMC10702970 | biostudies-literature
Project description:VerbNet-the most extensive online verb lexicon currently available for English-has proved useful in supporting a variety of NLP tasks. However, its exploitation in multilingual NLP has been limited by the fact that such classifications are available for few languages only. Since manual development of VerbNet is a major undertaking, researchers have recently translated VerbNet classes from English to other languages. However, no systematic investigation has been conducted into the applicability and accuracy of such a translation approach across different, typologically diverse languages. Our study is aimed at filling this gap. We develop a systematic method for translation of VerbNet classes from English to other languages which we first apply to Polish and subsequently to Croatian, Mandarin, Japanese, Italian, and Finnish. Our results on Polish demonstrate high translatability with all the classes (96% of English member verbs successfully translated into Polish) and strong inter-annotator agreement, revealing a promising degree of overlap in the resultant classifications. The results on other languages are equally promising. This demonstrates that VerbNet classes have strong cross-lingual potential and the proposed method could be applied to obtain gold standards for automatic verb classification in different languages. We make our annotation guidelines and the six language-specific verb classifications available with this paper.
| S-EPMC6428229 | biostudies-literature
Project description:The use of offensive terms in user-generated content on different social media platforms is one of the major concerns for these platforms. The offensive terms have a negative impact on individuals, which may lead towards the degradation of societal and civilized manners. The immense amount of content generated at a higher speed makes it humanly impossible to categorise and detect offensive terms. Besides, it is an open challenge for natural language processing (NLP) to detect such terminologies automatically. Substantial efforts are made for high-resource languages such as English. However, it becomes more challenging when dealing with resource-poor languages such as Urdu. Because of the lack of standard datasets and pre-processing tools for automatic offensive terms detection. This paper introduces a combinatorial pre-processing approach in developing a classification model for cross-platform (Twitter and YouTube) use. The approach uses datasets from two different platforms (Twitter and YouTube) the training and testing the model, which is trained to apply decision tree, random forest and naive Bayes algorithms. The proposed combinatorial pre-processing approach is applied to check how machine learning models behave with different combinations of standard pre-processing techniques for low-resource language in the cross-platform setting. The experimental results represent the effectiveness of the machine learning model over different subsets of traditional pre-processing approaches in building a classification model for automatic offensive terms detection for a low resource language, i.e., Urdu, in the cross-platform scenario. In the experiments, when dataset D1 is used for training and D2 is applied for testing, the pre-processing approach named Stopword removal produced better results with an accuracy of 83.27%. Whilst, in this case, when dataset D2 is used for training and D1 is applied for testing, stopword removal and punctuation removal were observed as a better preprocessing approach with an accuracy of 74.54%. The combinatorial approach proposed in this paper outperformed the benchmark for the considered datasets using classical as well as ensemble machine learning with an accuracy of 82.9% and 97.2% for dataset D1 and D2, respectively.
| S-EPMC10496005 | biostudies-literature
Project description:ObjectiveIn Hebrew online health communities, participants commonly write medical terms that appear as transliterated forms of a source term in English. Such transliterations introduce high variability in text and challenge text-analytics methods. To reduce their variability, medical terms must be normalized, such as linking them to Unified Medical Language System (UMLS) concepts. We present a method to identify both transliterated and translated Hebrew medical terms and link them with UMLS entities.Materials and methodsWe investigate the effect of linking terms in Camoni, a popular Israeli online health community in Hebrew. Our method, MDTEL (Medical Deep Transliteration Entity Linking), includes (1) an attention-based recurrent neural network encoder-decoder to transliterate words and mapping UMLS from English to Hebrew, (2) an unsupervised method for creating a transliteration dataset in any language without manually labeled data, and (3) an efficient way to identify and link medical entities in the Hebrew corpus to UMLS concepts, by producing a high-recall list of candidate medical terms in the corpus, and then filtering the candidates to relevant medical terms.ResultsWe carry out experiments on 3 disease-specific communities: diabetes, multiple sclerosis, and depression. MDTEL tagging and normalizing on Camoni posts achieved 99% accuracy, 92% recall, and 87% precision. When tagging and normalizing terms in queries from the Camoni search logs, UMLS-normalized queries improved search results in 46% of the cases.ConclusionsCross-lingual UMLS entity linking from Hebrew is possible and improves search performance across communities. Annotated datasets, annotation guidelines, and code are made available online (https://github.com/yonatanbitton/mdtel).
| S-EPMC7566404 | biostudies-literature
Project description:Automatic identification of offensive/abusive language is very necessary to get rid of unwanted behavior. However, it is more challenging to generalize the solution due to the different grammatical structures and vocabulary of each language. Most of the prior work targeted western languages, however, one study targeted a low-resource language (Urdu). The prior study used basic linguistic features and a small dataset. This study designed a new dataset (collected from popular Pakistani Facebook pages) containing 7,500 posts for offensive language detection in Urdu. The proposed methodology used four types of feature engineering models: three are frequency-based and the fourth one is the embedding model. Frequency-based are either determined by the term frequency-inverse document frequency (TF-IDF) or bag-of-words or word n-gram feature vectors. The fourth is generated by the word2vec model, trained on the Urdu embeddings using a corpus of 196,226 Facebook posts. The experiments demonstrate that the stacking-based ensemble model with word2vec shows the best performance as a standalone model by achieving 88.27% accuracy. In addition, the wrapper-based feature selection method further improves performance. The hybrid combination of TF-IDF, bag-of-words, and word2vec feature models achieved 90% accuracy and 97% AUC. In addition, it outperformed the baseline with an improvement of 3.55% in accuracy, 3.68% in the recall, 3.60% in f1-measure, 3.67% in precision, and 2.71% in AUC. The findings of this research provide practical implications for commercial applications and future research.
| S-EPMC10280260 | biostudies-literature
Project description:Efficient reading is essential for societal participation, so reading proficiency is a central educational goal. Here, we use an individualized diagnostics and training framework to investigate processes in visual word recognition and evaluate its usefulness for detecting training responders. We (i) motivated a training procedure based on the Lexical Categorization Model (LCM) to introduce the framework. The LCM describes pre-lexical orthographic processing implemented in the left-ventral occipital cortex and is vital to reading. German language learners trained their lexical categorization abilities while we monitored reading speed change. In three studies, most language learners increased their reading skills. Next, we (ii) estimated, for each word, the LCM-based features and assessed each reader's lexical categorization capabilities. Finally, we (iii) explored machine learning procedures to find the optimal feature selection and regression model to predict the benefit of the lexical categorization training for each individual. The best-performing pipeline increased reading speed from 23% in the unselected group to 43% in the machine-selected group. This selection process strongly depended on parameters associated with the LCM. Thus, training in lexical categorization can increase reading skills, and accurate computational descriptions of brain functions that allow the motivation of a training procedure combined with machine learning can be powerful for individualized reading training procedures.
| S-EPMC11006909 | biostudies-literature
Project description:Previous studies have shown that Chinese speakers and non-Chinese speakers exhibit different patterns of cross-modal congruence for the lexical tones of Mandarin Chinese, depending on which features of the pitch they attend to. But is this pattern of language-specific listening a conscious cultural strategy or an automatic processing effect? If automatic, does it also apply when the same pitch contours no longer sound like speech? Implicit Association Tests (IATs) provide an indirect measure of cross-modal association. In a series of IAT studies, conducted with participants with three kinds of language backgrounds (Chinese-dominant bilinguals, Chinese balanced bilinguals, and English speakers with no Chinese experience) we find language-specific congruence effects for Mandarin lexical tones but not for matched sine-wave stimuli. That is, for linguistic stimuli, non-Chinese speakers show advantages for pitch-height congruence (high-pointy, low-curvy); no congruence effects were found for Chinese speakers. For non-linguistic stimuli, all participant groups showed advantages for pitch-height congruence. The present findings suggest that non-lexical tone congruence (high-pointy, low-curvy) is a basic congruence pattern, and the acquisition of a language with lexical tone can alter this perception.
| S-EPMC10078355 | biostudies-literature
Project description:Detrimental to individuals and society, online hateful messages have recently become a major social issue. Among them, one new type of hateful message, "hateful meme", has emerged and brought difficulties in traditional deep learning-based detection. Because hateful memes were formatted with both text captions and images to express users' intents, they cannot be accurately identified by singularly analyzing embedded text captions or images. In order to effectively detect a hateful meme, the algorithm must possess strong vision and language fusion capability. In this study, we move closer to this goal by feeding a triplet by stacking the visual features, object tags, and text features of memes generated by the object detection model named Visual features in Vision-Language (VinVl) and the optical character recognition (OCR) technology into a Transformer-based Vision-Language Pre-Training Model (VL-PTM) OSCAR+ to perform the cross-modal learning of memes. After fine-tuning and connecting to a random forest (RF) classifier, our model (OSCAR+RF) achieved an average accuracy and AUROC of 0.684 and 0.768, respectively, on the hateful meme detection task in a public test set, which was higher than the other eleven (11) published baselines. In conclusion, this study has demonstrated that VL-PTMs with the addition of anchor points can improve the performance of deep learning-based detection of hateful memes by involving a more substantial alignment between the text caption and visual information.
| S-EPMC9467312 | biostudies-literature
Project description:Previous research has shown that familiarity with a talker's voice can improve linguistic processing (herein, "Familiar Talker Advantage"), but this benefit is constrained by the context in which the talker's voice is familiar. The current study examined how familiarity affects intelligibility by manipulating the type of talker information available to listeners. One group of listeners learned to identify bilingual talkers' voices from English words, where they learned language-specific talker information. A second group of listeners learned the same talkers from German words, and thus only learned language-independent talker information. After voice training, both groups of listeners completed a word recognition task with English words produced by both familiar and unfamiliar talkers. Results revealed that English-trained listeners perceived more phonemes correct for familiar than unfamiliar talkers, while German-trained listeners did not show improved intelligibility for familiar talkers. The absence of a processing advantage in speech intelligibility for the German-trained listeners demonstrates limitations on the Familiar Talker Advantage, which crucially depends on the language context in which the talkers' voices were learned; knowledge of how a talker produces linguistically relevant contrasts in a particular language is necessary to increase speech intelligibility for words produced by familiar talkers.
| S-EPMC3253604 | biostudies-other
Project description:Bilingual dictionaries for technical terms such as biomedical terms are an important resource for machine translation systems as well as for humans who would like to understand a concept described in a foreign language. Often a biomedical term is first proposed in English and later it is manually translated to other languages. Despite the fact that there are large monolingual lexicons of biomedical terms, only a fraction of those term lexicons are translated to other languages. Manually compiling large-scale bilingual dictionaries for technical domains is a challenging task because it is difficult to find a sufficiently large number of bilingual experts. We propose a cross-lingual similarity measure for detecting most similar translation candidates for a biomedical term specified in one language (source) from another language (target). Specifically, a biomedical term in a language is represented using two types of features: (a) intrinsic features that consist of character n-grams extracted from the term under consideration, and (b) extrinsic features that consist of unigrams and bigrams extracted from the contextual windows surrounding the term under consideration. We propose a cross-lingual similarity measure using each of those feature types. First, to reduce the dimensionality of the feature space in each language, we propose prototype vector projection (PVP)--a non-negative lower-dimensional vector projection method. Second, we propose a method to learn a mapping between the feature spaces in the source and target language using partial least squares regression (PLSR). The proposed method requires only a small number of training instances to learn a cross-lingual similarity measure. The proposed PVP method outperforms popular dimensionality reduction methods such as the singular value decomposition (SVD) and non-negative matrix factorization (NMF) in a nearest neighbor prediction task. Moreover, our experimental results covering several language pairs such as English-French, English-Spanish, English-Greek, and English-Japanese show that the proposed method outperforms several other feature projection methods in biomedical term translation prediction tasks.
| S-EPMC4452086 | biostudies-literature