Controllable Text Generation for Open-Domain Creativity and Fairness

Abstract

The demand for machine learning systems that can provide both transparency and fairness is constantly growing. Since the concept of fairness depends on the context, studies in the literature have proposed various formalisation and mitigation strategies. In this work, we propose a novel, flexible, discrimination-aware classifier that allows the user to: (i) select and mitigate the desired fairness criterion from a set of available options; (ii) implement more than one fairness criterion; (iii) handle more than one sensitive attribute; and (iv) specify the desired level of fairness to meet specific business needs or regulatory requirements. Our approach is based on an optimised extension to the decision-tree classifier, and aims to provide transparent and fair rules to the final users.

Toward Unified Controllable Text Generation via Regular Expression Instruction

“…We compare the performance of our method with other baselines, including fine-tuning methods BART (Lin et al, 2020) and T5-Large (Lin et al, 2020), auxilary guiding model method NADO (Meng et al, 2022), prompting method NRP (Carlsson et al, 2022), and 8-shot pure natural language instruction (NLI) on GPT-3.5, which is shown at Table 3a.…”

“…Constrainted Beam Search (Anderson et al, 2017), DeLorean (Qin et al, 2020), COLD , Neuro-Logic (Lu et al, 2021); or using auxilary guiding model, e.g. PPLM (Anderson et al, 2017), GeDI (Krause et al, 2021), FUDGE (Yang and Klein, 2021), CTRLsum (He et al, 2022), Plug-and-Play Content Planning , NADO (Meng et al, 2022), and MACSum (Zhang et al, 2023) .…”

“…For controllable text generation, typical fine-grained control tasks include lexicon (Lin et al, 2020), generating position and length (Carlsson et al, 2022). Recently, various approaches have been proposed to satisfy these constraints, which can be categorized into three different paradigms: retraining or refactoring the model (Keskar et al, 2019;He, 2021;Chan et al, 2021a); tuning on given data (Lester et al, 2021;Stiennon et al, 2020a); manually designed post-processing (Qin et al, 2020Meng et al, 2022;Lu et al, 2021.…”

Evaluating Large Language Models on Controlled Generation Tasks

“…How to Improve. We encourage future work to explore from two different directions: 1) chain/tree/graph-of-thought reasoning, and 2) bridging LLMs with non-autoregressive generation abilities (e.g., NADO (Meng et al, 2022)). For the first one, one can try both simple chain/tree/graphof-thought prompting or even pretrained LLMs with chain-of-thought/scratchpad pairs, as it shows promises for mathematical reasoning .…”

Harnessing Black-Box Control to Boost Commonsense in LM’s Generation

“…In this work, we propose BOOST, a framework to boost the commonsense of PLTMs' generation in a plug-and-play manner (Figure 2), which is inspired by the recent development of controllable generation to use a small auxiliary model to control a PTLM by training on its self-generated samples (Meng et al, 2022). Specifically, to better integrate commonsense knowledge, we first build a scorer that evaluates how commonsensical a sentence is.…”