Models for detecting toxic content play an important role in keeping people safe online. There has been much progress in detecting overt toxicity. Covert toxicity, however, remains a challenge because its detection requires an understanding of implicit meaning and subtle connotations. In this paper, we explore the potential of leveraging references, such as external knowledge and textual interpretations, to enhance the detection of covert toxicity. We run experiments on two covert toxicity datasets with two types of references: 1) information retrieved from a search API, and 2) interpretations generated by large language models. We find that both types of references improve detection, with the latter being more useful than the former. We also find that generating interpretations grounded on properties of covert toxicity, such as humor and irony, lead to the largest improvements

Toxic language, such as hate speech, can deter users from participating in online communities and enjoying popular platforms. Previous approaches to detecting toxic language and norm violations have been primarily concerned with conversations from online forums and social media, such as Reddit and Twitter. These approaches are less effective when applied to conversations on live-streaming platforms, such as Twitch and YouTube Live, as each comment is only visible for a limited time and lacks a thread structure that establishes its relationship with other comments. In this work, we share the first NLP study dedicated to detecting norm violations in conversations on live-streaming platforms. We define norm violation categories in live-stream chats and annotate 4,583 moderated comments from Twitch. We articulate several facets of live-stream data that differ from other forums, and demonstrate that existing models perform poorly in this setting. By conducting a user study, we identify the informational context humans use in live-stream moderation, and train models leveraging context to identify norm violations. Our results show that appropriate contextual information can boost moderation performance by 35%.

Recent directions for offensive language detection are hierarchical modeling, identifying the type and the target of offensive language, and interpretability with offensive span annotation and prediction. These improvements are focused on English and do not transfer well to other languages because of cultural and linguistic differences. In this paper, we present the Korean Offensive Language Dataset (KOLD) comprising 40,429 comments, which are annotated hierarchically with the type and the target of offensive language, accompanied by annotations of the corresponding text spans. We collect the comments from NAVER news and YouTube platform and provide the titles of the articles and videos as the context information for the annotation process. We use these annotated comments as training data for Korean BERT and RoBERTa models and find that they are effective at offensiveness detection, target classification, and target span detection while having room for improvement for target group classification and offensive span detection. We discover that the target group distribution differs drastically from the existing English datasets, and observe that providing the context information improves the model performance in offensiveness detection (+0.3), target classification (+1.5), and target group classification (+13.1). We publicly release the dataset and baseline models.

In this technical report, we describe the fine-tuned ASR-MT pipeline used for the IWSLT shared task. We remove less useful speech samples by checking WER with an ASR model, and further train a wav2vec and Transformers-based ASR module based on the filtered data. In addition, we cleanse the errata that can interfere with the machine translation process and use it for Transformer-based MT module training. Finally, in the actual inference phase, we use a sentence boundary detection model trained with constrained data to properly merge fragment ASR outputs into full sentences. The merged sentences are post-processed using part of speech. The final result is yielded by the trained MT module. The performance using the dev set displays BLEU 20.37, and this model records the performance of BLEU 20.9 with the test set.

Social consensus has been established on the severity of online hate speech since it not only causes mental harm to the target, but also gives displeasure to the people who read it. For Korean, the definition and scope of hate speech have been discussed widely in researches, but such considerations were hardly extended to the construction of hate speech corpus. Therefore, we create a Korean online hate speech dataset with concrete annotation guideline to see how real world toxic expressions concern sociolinguistic discussions. This inductive observation reveals that hate speech in online news comments is mainly composed of social bias and toxicity. Furthermore, we check how the final corpus corresponds with the definition and scope of hate speech, and confirm that the overall procedure and outcome is in concurrence with the sociolinguistic discussions.

This paper describes the system submitted by Papago team for the quality estimation task at WMT 2020. It proposes two key strategies for quality estimation: (1) task-specific pretraining scheme, and (2) task-specific data augmentation. The former focuses on devising learning signals for pretraining that are closely related to the downstream task. We also present data augmentation techniques that simulate the varying levels of errors that the downstream dataset may contain. Thus, our PATQUEST models are exposed to erroneous translations in both stages of task-specific pretraining and finetuning, effectively enhancing their generalization capability. Our submitted models achieve significant improvement over the baselines for Task 1 (Sentence-Level Direct Assessment; EN-DE only), and Task 3 (Document-Level Score).

Toxic comments in online platforms are an unavoidable social issue under the cloak of anonymity. Hate speech detection has been actively done for languages such as English, German, or Italian, where manually labeled corpus has been released. In this work, we first present 9.4K manually labeled entertainment news comments for identifying Korean toxic speech, collected from a widely used online news platform in Korea. The comments are annotated regarding social bias and hate speech since both aspects are correlated. The inter-annotator agreement Krippendorff’s alpha score is 0.492 and 0.496, respectively. We provide benchmarks using CharCNN, BiLSTM, and BERT, where BERT achieves the highest score on all tasks. The models generally display better performance on bias identification, since the hate speech detection is a more subjective issue. Additionally, when BERT is trained with bias label for hate speech detection, the prediction score increases, implying that bias and hate are intertwined. We make our dataset publicly available and open competitions with the corpus and benchmarks.

Quality estimation (QE), a task of evaluating the quality of translation automatically without human-translated reference, is one of the important challenges for machine translation (MT). As the QE methods, BLEU score for round-trip translation (RTT) had been considered. However, it was found to be a poor predictor of translation quality since BLEU was not an adequate metric to detect semantic similarity between input and RTT. Recently, the pre-trained language models have made breakthroughs in many NLP tasks by providing semantically meaningful word and sentence embeddings. In this paper, we employ the semantic embeddings to RTT-based QE metric. Our method achieves the highest correlations with human judgments compared to WMT 2019 quality estimation metric task submissions. Additionally, we observe that with semantic-level metrics, RTT-based QE is robust to the choice of a backward translation system and shows consistent performance on both SMT and NMT forward translation systems.