English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Warning: this paper contains model outputs exhibiting offensiveness andbiases. Recently pre-trained language models (PLMs) have prospered in variousnatural language generation (NLG) tasks due to their ability to generate fairlyfluent text. Nevertheless, these models are observed to capture and reproduceharmful contents in training corpora, typically toxic language and socialbiases, raising severe moral issues. Prior works on ethical NLG tackledetoxifying and debiasing separately, which is problematic since we finddebiased models still exhibit toxicity while detoxified ones even exacerbatesocial biases. To address such a challenge, we propose the first unifiedframework of detoxifying and debiasing called UDDIA, which jointly formalizesthese two problems as rectifying the output space. We theoretically interpretour framework as learning a text distribution mixing weighted attributes.Besides, UDDIA conducts adaptive optimization of only a few parameters duringdecoding based on a parameter-efficient tuning schema without any trainingdata. This leads to minimal generation quality loss and improved rectificationperformance with acceptable computational cost. Experimental resultsdemonstrate that compared to several strong baselines, UDDIA achieves debiasingand detoxifying simultaneously and better balances efficiency andeffectiveness, taking a further step towards practical ethical NLG.

Unified Detoxifying and Debiasing in Language Generation via Inference-time Adaptive Optimization