2026-09-20T07:56:31Z https://www.ideals.illinois.edu/oai-pmh

oai:www.ideals.illinois.edu:2142/120261 2023-09-05 col_2142_5131 col_2142_10761 com_2142_5130 com_2142_10755 com_2142_234

Sundaram, Hari 2023-05 application/pdf en text Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2023-09-01 without embargo terms The student, Samraj Moorjani, accepted the attached license on 2023-04-12 at 17:34. The student, Samraj Moorjani, submitted this Thesis for approval on 2023-04-12 at 17:35. This Thesis was approved for publication on 2023-04-14 at 13:25. DSpace SAF Submission Ingestion Package generated from Vireo submission #18970 on 2023-09-01 at 17:08:31 Controllable natural language generation for audience-centric styles Moorjani, Samraj 2023-04-14 Natural Langauge Generation Persuasiveness Audience-centric Styles Persuasive Natural Language Generation Adopting contextually appropriate, audience-tailored linguistic styles, namely persuasiveness, is critical to the success of user-centric language generation systems (e.g., chatbots, computer-aided writing, dialog systems). While existing approaches demonstrate textual style transfer with large volumes of data, grounding style on audience-independent factors is innately limiting because many stylistic objectives (e.g., persuasiveness, memorability, empathy) are hard to define without audience feedback. In this thesis, we first propose the novel task of style infusion - infusing the stylistic preferences of audiences in pretrained language generation models. Since humans are better at pairwise comparisons than direct scoring - i.e., is Sample-A more persuasive than SampleB - we leverage limited pairwise human judgments to bootstrap a style analysis model and augment our seed set of judgments. We infuse the learned textual style in a GPT-2 based text generator while balancing fluency and style adoption. With quantitative and novel qualitative assessments, we show that our infusion approach can generate compelling stylized examples with generic text prompts. We then utilize complex linguistic features strongly correlated with persuasiveness to guide the generation of sequence-to-sequence models. We explore modifications of two approaches - an edit-then-prototype model and the style infusion architecture - to exhibit a “tuning-knob”- like control over the speed of text (i.e., how quickly content is covered). We empirically show that our modifications lead to strong controls over generated text and discuss directions to improve the fluency and control of generations further. Thesis eng https://hdl.handle.net/2142/120261 Copyright 2023 Samraj Moorjani M.S. Thesis Computer Science University of Illinois at Urbana-Champaign Computer Science