Language is constantly changing and evolving, leaving language models to quickly become outdated, both factually and linguistically. Recent research proposes we continuously update our models using new data. Continuous training allows us to teach language models about new events and facts and changing norms. However, continuous training also means continuous costs. We show there is currently limited evidence for the benefits of continuous training, be it for the actual downstream performance or the environmental cost. Our results show continuous training does not significantly improve performance. While it is clear that, sooner or later, our language models need to be updated, it is unclear when this effort is worth the cost. We call for a critical reflection about when and how to use continuous training and for more benchmarks to support this research direction.
Many interpretability tools allow practitioners and researchers to explain Natural Language Processing systems. However, each tool requires different configurations and provides explanations in different forms, hindering the possibility of assessing and comparing them. A principled, unified evaluation benchmark will guide the users through the central question: which explanation method is more reliable for my use case? We introduce ferret, an easy-to-use, extensible Python library to explain Transformer-based models integrated with the Hugging Face Hub. It offers a unified benchmarking suite to test and compare a wide range of state-of-the-art explainers on any text or interpretability corpora. In addition, ferret provides convenient programming abstractions to foster the introduction of new explanation methods, datasets, or evaluation metrics.
Hate speech detection models are typically evaluated on held-out test sets. However, this risks painting an incomplete and potentially misleading picture of model performance because of increasingly well-documented systematic gaps and biases in hate speech datasets. To enable more targeted diagnostic insights, recent research has thus introduced functional tests for hate speech detection models. However, these tests currently only exist for English-language content, which means that they cannot support the development of more effective models in other languages spoken by billions across the world. To help address this issue, we introduce Multilingual HateCheck (MHC), a suite of functional tests for multilingual hate speech detection models. MHC covers 34 functionalities across ten languages, which is more languages than any other hate speech dataset. To illustrate MHC's utility, we train and test a high-performing multilingual hate speech detection model, and reveal critical model weaknesses for monolingual and cross-lingual applications.
Natural Language Processing (NLP) models risk overfitting to specific terms in the training data, thereby reducing their performance, fairness, and generalizability. E.g., neural hate speech detection models are strongly influenced by identity terms like gay, or women, resulting in false positives, severe unintended bias, and lower performance. Most mitigation techniques use lists of identity terms or samples from the target domain during training. However, this approach requires a-priori knowledge and introduces further bias if important terms are neglected. Instead, we propose a knowledge-free Entropy-based Attention Regularization (EAR) to discourage overfitting to training-specific terms. An additional objective function penalizes tokens with low self-attention entropy. We fine-tune BERT via EAR: the resulting model matches or exceeds state-of-the-art performance for hate speech classification and bias metrics on three benchmark corpora in English and Italian. EAR also reveals overfitting terms, i.e., terms most likely to induce bias, to help identify their effect on the model, task, and predictions.
Meaning is context-dependent, but many properties of language (should) remain the same even if we transform the context. For example, sentiment, entailment, or speaker properties should be the same in a translation and original of a text. We introduce language invariant properties: i.e., properties that should not change when we transform text, and how they can be used to quantitatively evaluate the robustness of transformation algorithms. We use translation and paraphrasing as transformation examples, but our findings apply more broadly to any transformation. Our results indicate that many NLP transformations change properties like author characteristics, i.e., make them sound more male. We believe that studying these properties will allow NLP to address both social factors and pragmatic aspects of language. We also release an application suite that can be used to evaluate the invariance of transformation applications.
Many data sets in a domain (reviews, forums, news, etc.) exist in parallel languages. They all cover the same content, but the linguistic differences make it impossible to use traditional, bag-of-word-based topic models. Models have to be either single-language or suffer from a huge, but extremely sparse vocabulary. Both issues can be addressed by transfer learning. In this paper, we introduce a zero-shot cross-lingual topic model, i.e., our model learns topics on one language (here, English), and predicts them for documents in other languages. By using the text of the same document in different languages, we can evaluate the quality of the predictions. Our results show that topics are coherent and stable across languages, which suggests exciting future research directions.
Recently, Natural Language Processing (NLP) has witnessed an impressive progress in many areas, due to the advent of novel, pretrained contextual representation models. In particular, Devlin et al. (2019) proposed a model, called BERT (Bidirectional Encoder Representations from Transformers), which enables researchers to obtain state-of-the art performance on numerous NLP tasks by fine-tuning the representations on their data set and task, without the need for developing and training highly-specific architectures. The authors also released multilingual BERT (mBERT), a model trained on a corpus of 104 languages, which can serve as a universal language model. This model obtained impressive results on a zero-shot cross-lingual natural inference task. Driven by the potential of BERT models, the NLP community has started to investigate and generate an abundant number of BERT models that are trained on a particular language, and tested on a specific data domain and task. This allows us to evaluate the true potential of mBERT as a universal language model, by comparing it to the performance of these more specific models. This paper presents the current state of the art in language-specific BERT models, providing an overall picture with respect to different dimensions (i.e. architectures, data domains, and tasks). Our aim is to provide an immediate and straightforward overview of the commonalities and differences between Language-Specific (language-specific) BERT models and mBERT. We also provide an interactive and constantly updated website that can be used to explore the information we have collected, at https://bertlang.unibocconi.it.