TheNorth @ HaSpeeDe 2: BERT-based Language Model Fine-tuning for Italian Hate Speech Detection

Abstract

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 Englishto-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.0 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. * Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started the effort to evaluate this idea. Ashish, with Illia, designed and implemented the first Transformer models and has been crucially involved in every aspect of this work. Noam proposed scaled dot-product attention, multi-head attention and the parameter-free position representation and became the other person involved in nearly every detail. Niki designed, implemented, tuned and evaluated countless model variants in our original codebase and tensor2tensor. Llion also experimented with novel model variants, was responsible for our initial codebase, and efficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and implementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating our research.† Work performed while at Google Brain. ‡ Work performed while at Google Research.

HATE-ITA: Hate Speech Detection in Italian Social Media Text

“…National evaluation campaigns and shared tasks played a significant role in releasing non-English corpora for hate speech detection (Wiegand et al, 2018;Mulki and Ghanem, 2021;Basile et al, 2019;Ptaszynski et al, 2019). Indeed, the research of hate speech detection in Italian in mono-lingual settings mainly revolves around the datasets (Fersini et al, 2018;Sanguinetti et al, 2020;Fersini et al, 2020b) released for shared tasks (Bakarov, 2018;Cimino et al, 2018;Attanasio and Pastor, 2020;Lees et al, 2020;Lavergne et al, 2020;Fersini et al, 2020a;Attanasio et al, 2022a, inter alia).…”

Proceedings of the Sixth Workshop on Online Abuse and Harms (WOAH)

“…National evaluation campaigns and shared tasks played a significant role in releasing non-English corpora for hate speech detection . Indeed, the research of hate speech detection in Italian in mono-lingual settings mainly revolves around the datasets Fersini et al, 2020b) released for shared tasks (Bakarov, 2018;Cimino et al, 2018;Attanasio and Pastor, 2020;Lees et al, 2020;Lavergne et al, 2020;Fersini et al, 2020a;Attanasio et al, 2022a, inter alia).…”

Sarcasm and Implicitness in Abusive Language Detection: A Multilingual Perspective

“…To our knowledge, only few scholars experimented on the contribution of less explicit information, such as metaphors [Lemmens et al, 2021] and stereotypes [Lavergne et al, 2020], for abusive language detection. This motivated us to investigate more about how different dimensions of hate interact among them.…”

“…Interestingly, Lemmens et al [2021] proved the contribution of hateful metaphors as features for the identification of the type and target of hate speech in Dutch Facebook comments in models based on classical machine learning and transformers. Whereas Lavergne et al [2020] exploit the multi-annotation proposed in HaSpeeDe2020 about the presence of hate speech and stereotype in tweets to train a multi-task learning-based model reaching the best score in hate speech detection in tweets.…”

“…TheNorth Lavergne et al [2020] TheNorth team experimented the fine-tuning of various versions of BERT (such as AlBERTo , UmBERTo 45 and so on) for both runs. In particular, they used a linear layer with a softmax on top of the CLS token, applying a novel technique of layer-wise learning rate.…”