Kai-Wei Chang

TL;DR: Experiments on modified Word Embedding Association Test, word similarity, word translation, and word pair translation tasks show that the proposed approaches can effectively reduce the gender bias while preserving the utility of the original embeddings.

TL;DR: A novel framework, learning to discriminate perturbation (DISP), to identify and adjust malicious perturbations, thereby blocking adversarial attacks for text classification models and shows the robustness of DISP across different situations.

TL;DR: The Bias in Open-Ended Language Generation Dataset (BOLD) as mentioned in this paper is a large-scale dataset that consists of 23,679 English text generation prompts for bias benchmarking across five domains: profession, gender, race, religion and political ideology.

TL;DR: This paper developed a method to induce societal biases in generated text when input prompts contain mentions of specific demographic groups, and analyzed two scenarios: 1) inducing negative biases for one demographic and positive biases for another demographic, and 2) equalizing biases between demographics.

TL;DR: Investigating crosslingual transfer and posit that an orderagnostic model will perform better when transferring to distant foreign languages shows that RNN-based architectures transfer well to languages that are close to English, while self-attentive models have better overall cross-lingualtransferability and perform especially well on distant languages.