Korea, Republic of

Standard approaches to hate speech detection rely on sufficient available hate speech annotations. Extending previous work that repurposes natural language inference (NLI) models for zero-shot text classification, we propose a simple approach that combines multiple hypotheses to improve English NLI-based zero-shot hate speech detection. We first conduct an error analysis for vanilla NLI-based zero-shot hate speech detection and then develop four strategies based on this analysis. The strategies use multiple hypotheses to predict various aspects of an input text and combine these predictions into a final verdict. We find that the zero-shot baseline used for the initial error analysis already outperforms commercial systems and fine-tuned BERT-based hate speech detection models on HateCheck. The combination of the proposed strategies further increases the zero-shot accuracy of 79.4% on HateCheck by 7.9 percentage points (pp), and the accuracy of 69.6% on ETHOS by 10.0pp.

COLING 2022

Hypothesis Engineering for Zero-Shot and Few-Shot Hate Speech Detection

toxic language

abusive language

hate speech

zero-shot

[![](https://assets.underline.io/uploads/markdown_image/1/image/81a3c0317d24f663b49b996875024d45.png)](https://aclanthology.org/volumes/2022.coling-1/)



### THE CONFERENCE WE KNOW AND WE WANT.
**COLING**, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

First established in 1965, the biennial COLING conference is held in diverse parts of the globe and attracts participants from both top-ranked research centers and emerging countries. Today, the most important developments in our field are taking place not only in universities and academic research institutes but also in industrial research departments including tech-startups. COLING provides opportunities for all these communities to showcase their exciting discovery.

In fall of 2022, COLING will be held in Gyeongju in a hybrid format. All participants can either present at the venue site or join virtually. As more people get vaccinated, we are happy to provide safer environments for our colleagues. We believe that COLING 2022 will be one of the conferences, free from the pandemic. The hybrid format gives presenters and sponsors a valuable opportunity to promote their companies in both an online and in-person venue. For the first time in a long time, customers can interact with their sponsor's products first-hand. The online venue, too, gives sponsors the chance to network with those unable to attend the in-person session.

The hybrid format ultimately lends itself to greater exposure for our sponsors: COLING2022 will let you reach more potential partners and customers than ever before!<br><br><br>

To gain access to this event page you are required to register.<br> Find more information and pay the registration fee on event the organizer’s website **[https://coling2022.org/reg](https://coling2022.org/reg)**

COLING, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

workshop paper

In this presentation, we discuss an interpretable frame-
work to integrate toxic language annotations.
Most data sets address only one aspect of the
complex relationship in toxic communication
and are inconsistent with each other. Enrich-
ing annotations with more details and infor-
mation is, however, of great importance in or-
der to develop high-performing and compre-
hensive explainable language models. Such
systems should recognize and interpret both
expressions that are toxic as well as expres-
sions that make reference to specific targets to
combat toxic language. We, therefore, create
a crowd-annotation task to mark the spans of
words that refer to target communities as an ex-
tension of the HateXplain data set. We present
a quantitative and qualitative analysis of the
annotations. We also fine-tune RoBERTa-base
on our data and experiment with different USQ (Media Unit Quality Score)
thresholds to measure their effect on the clas-
sification. The F1-score of our best model on
the test set is 79%. The annotations are freely
available and can be combined with the exist-
ing HateXplain annotations to build richer and
more complete models.

Annotating targets of toxic language at the span level

Annotating abusive language is expensive, logistically complex and creates a risk of psychological harm. However, most machine learning research has prioritized maximizing effectiveness (i.e., F1 or accuracy score) rather than data efficiency (i.e., minimizing the amount of data that is annotated). In this paper, we use simulated experiments over two datasets at varying percentages of abuse to demonstrate that transformers-based active learning is a promising approach to substantially raise efficiency whilst still maintaining high effectiveness, especially when abusive content is a smaller percentage of the dataset. This approach requires a fraction of labeled data to reach performance equivalent to training over the full dataset.

Is More Data Better? Re-thinking the Importance of Efficiency in Abusive Language Detection with Transformers-Based Active Learning

Highly imbalanced textual datasets continue to pose a challenge for supervised learning models. However, viewing such imbalanced text data as an anomaly detection (AD) problem has advantages for certain tasks such as detecting hate speech, or inappropriate and/or offensive language in large social media feeds. There the unwanted content tends to be both rare and non-uniform with respect to its thematic character, and better fits the definition of an anomaly than a class. Several recent approaches to textual AD use transformer models, achieving good results but with trade-offs in pre-training and inflexibility with respect to new domains. In this paper we compare two linear models within the NMF family, which also have a recent history in textual AD. We introduce a new approach based on an alternative regularization of the NMF objective. Our results surpass other linear AD models and are on par with deep models, performing comparably well even in very small outlier concentrations.

A Lightweight yet Robust Approach to Textual Anomaly Detection

The proliferation of online hate speech has necessitated the creation of algorithms which can detect toxicity. Most of the past research focuses on this detection as a classification task, but assigning an absolute toxicity label is often tricky. Hence, few of the past works transform the same task into a regression. This paper shows the comparative evaluation of different transformers and traditional machine learning
models on a recently released toxicity severity measurement dataset by Jigsaw. We further demonstrate the issues with the model predictions using explainability analysis

Which one is more toxic? Findings from Jigsaw Rate Severity of Toxic Comments

Automated textual cyberbullying detection is known to be a challenging task. It is sometimes expected that messages associated with bullying will either be a) busive, b) targeted at a specific individual or group, or c) have a negative sentiment. Transfer learning by fine-tuning pre-trained attention-based transformer language models (LMs) has achieved near state-of-the-art (SOA) precision in identifying textual fragments as being bullying-related or not. This study looks closely at two SOA LMs, BERT and HateBERT, fine-tuned on real-life cyberbullying datasets from multiple social networking platforms. We intend to determine whether these finely calibrated pre-trained LMs learn textual cyberbullying attributes or syntactical features in the text. The results of our comprehensive experiments show that despite the fact that attention weights are drawn more strongly to syntactical features of the text at every layer, attention weights cannot completely account for the decision-making of such attention-based transformers.

Can attention-based transformers explain or interpret cyberbullying detection?

Online hate speech detection is an inherently challenging task that has recently received much attention from the natural language processing community. Despite a substantial increase in performance, considerable challenges remain and include encoding contextual information into automated hate speech detection systems. In this paper, we focus on detecting the target of hate speech in Dutch social media: whether a hateful Facebook comment is directed against migrants or not (i.e., against someone else). We manually annotate the relevant conversational context and investigate the effect of different aspects of context on performance when adding it to a Dutch transformer-based pre-trained language model, BERTje. We show that performance of the model can be significantly improved by integrating relevant contextual information.

The Role of Context in Detecting the Target of Hate Speech

Information extraction on documents still remains a challenge, especially when dealing with unstructured documents with complex and variable layouts. Graph Neural Networks seem to be a promising approach to overcome these difficulties due to their flexible and sparse nature, but they have not been exploited yet. In this work, we present a multi-level graph-based model that performs entity building and linking on unstructured documents, purely based on GNNs, and extremely light (0.3 million parameters). We also propose a novel strategy for an optimal propagation of the information between the graph levels based on hypernodes. The conducted experiments on public and private datasets demonstrate that our model is suitable for solving the tasks, and that the proposed propagation strategy is optimal and outperforms other approaches.

Multilevel Hypernode Graphs for Effective and Efficient Entity Linking

A Knowledge Graph (KG) is the directed graphical representation of entities and relations in the real world. KG can be applied in diverse Natural Language Processing (NLP) tasks where knowledge is required. The need to scale up and complete KG automatically yields Knowledge Graph Embedding (KGE), a shallow machine learning model that is suffering from memory and training time consumption issues. To mitigate the computational load, we propose a parameter-sharing method, i.e., using conjugate parameters for complex numbers employed in KGE models. Our method improves memory efficiency by 2x in relation embedding while achieving comparable performance to the state-of-the-art non-conjugate models, with faster, or at least comparable, training time. We demonstrated the generalizability of our method on two best-performing KGE models $5^{\bigstar}\mathrm{E}$ \cite{Nay21} and $\mathrm{ComplEx}$ \cite{Tro16} on five benchmark datasets.

Sharing Parameter by Conjugation for Knowledge Graph Embeddings in Complex Space

The present paper introduces an ongoing research which aims to detect interpretable adjectival senses from monolingual corpora applying an unsupervised WSI approach. According to our expectations the findings of our investigation are going to contribute to the work of lexicographers, linguists and also facilitate the creation of benchmarks with semantic information for the NLP community. For doing so, we set up four criteria to distinguish between senses. We experiment with a graphical approach to model our criteria and then perform a detailed, linguistically motivated manual evaluation of the results.

A Clique-Based Graphical Approach to Detect Interpretable Adjectival Senses in Hungarian

Natural Language Premise Selection (NLPS) is a mathematical Natural Language Processing (NLP) task that retrieves a set of useful relevant premises to support the end-user finding the proof for a particular statement. In this paper, we evaluate the impact of Transformer-based contextual information and different fundamental similarity scores toward NLPS. The results demonstrate that the contextual representation is better at capturing meaningful information despite not being pretrained on mathematical background in comparison with the statistical approach (e.g., the TF-IDF) with a boost of around 3.00% MAP@500.

Premium content

Hypothesis Engineering for Zero-Shot and Few-Shot Hate Speech Detection

Downloads

Next from COLING 2022

Annotating targets of toxic language at the span level

Similar lecture

The Role of Context in Detecting the Target of Hate Speech