abstract: This dissertation theorizes Bad Faith Rhetorics, or, rhetorical gestures that work to derail, block, or otherwise stymy knowledge-building efforts. This work explores the ways that interventions against existing social hierarchies (i.e., feminist and antiracist interventions) build knowledge (that is, are epistemologically active), and the ways that bad faith rhetorics derail such interventions. This dissertation demonstrates how bad faith rhetorics function to defend the status quo, with its social stratification by race, gender, class, and other intersectional axes of identity. Bad faith argumentative maneuvers are abundant in online environments. Consequently, this dissertation offers two case studies of the comment sections of two TED Talks: Mellody Hobson’s “Color Blind or Color Brave?” and Juno Mac’s “The Laws that Sex Workers Really Want.” The central analyses deploy online ethnographic field methods and close reading to characterize bad faith rhetorical responses and to identify 1.) trends in such responses, 2.) the net effects on other conversational participants, and 3.) bad faith rhetoric mitigation strategies. This work engages Sartre’s work on Bad Faith, rhetoric scholarship on the knowledge-building affordances of argument, public sphere theory, critical race studies, and feminist scholarship. This dissertation’s theorization and case studies illustrate the pitfalls of specific counterproductive argumentative tactics that block progress toward more equitable ways of being (bad faith rhetorics), and makes several preliminary recommendations for mitigating such moves. / Dissertation/Thesis / Doctoral Dissertation English 2019

Metadiscourse, or discourse about discourse, manifests itself in both spoken and written genres. However, the focus of existing studies of spoken metadiscourse remains mainly on academic speaking (Hyland, 2017), whereas research on metadiscourse in spoken non-academic genres is much less common. The present study thus investigates, both qualitatively and quantitatively, the use of metadiscourse in TED (Technology, Entertainment, Design) talks, which are primarily monologic speeches aiming at knowledge dissemination among a lay audience. The corpus for this exploratory study comprises the top 20 most viewed speeches (49,000 tokens and 286 minutes of presentation time) presented in English at an official TED conference in 2019. This study attempts at shedding light on how TED speakers use metadiscourse from the perspective of the discourse functions by applying Ädel’s (2023) taxonomy of reflexive metadiscourse. The results reveal that TED speakers demonstrate a high level of audience orientation. The findings also indicate a strong focus on discourse organization, which underlines the importance of making this explicit in TED talks, thus providing guidance for the audience. One of the key observations of this study is that the audience in the genre of TED talks is complex, as talks are not only presented live when recorded, but also made available to a global, online audience. The thesis aims to contribute insights to TED talks as a genre, enhance the taxonomy development for metadiscourse, and suggest potential pedagogical implications for teaching public speaking.

要設計一場具影響力的公眾演說，演講者不能忽略演講情境的存在；要了解演講者如何進行演講設計，就須從其面臨之情境著手。過往以公眾演講為研究主題的文獻與坊間書籍，多聚焦於演講者如何運用演說技巧，或分析知名演講者進行演講時所使用的特定規則及方法，雖可窺探演講者如何進行演說設計，但對於情境的存在鮮少論及，亦無法真正擷取演講者為何如此規畫的原由。再者，演講之中的情境來自於四面八方，發掘演講者如何篩選及妥適運用情境，才是對於想要培養及建立公眾演講技巧者的最佳學習途徑。因此，本研究與六名TEDxTaipei演講者進行深度訪談，從演講者觀點出發，分析演講者感知哪些情境、並如何因應情境設計演講內容。 / To design a powerful public speech, a speaker cannot neglect the existence of speech contexts. To understand how a speaker designs his/her speech, we should first look at the context that the speaker faces. The majority of the literature reviews and books tend to focus on the public speaking skills of the speaker or analyze specific rules or methods adopted by famous speakers. Although this offers a glimpse into how speakers design their speeches, it seldom discusses the existence of speech contexts, nor is it able to pinpoint the reasons why a speaker takes a certain approach. Furthermore, since speech contexts vary in different places, guiding speakers how to select and make use of contexts is the best teaching approach. Therefore, this study conducted in-depth interviews with six speakers from TEDxTaipei in order to analyze what context they are aware of and how they design their speeches based on such context.

Have you ever listened to a speech, seriously attempted to discern the speaker’s message, then realize you have forgotten most of, if not, the entire message moments after the speech is finished? Far too often audiences sit through a presentation focusing as best as they can, only to have the speaker craft a message in a way that is nearly impossible for the audience to remember. The best speakers not only deliver a memorable message, but also one that inspires their audience to action or change of mindset. Speakers at Ted Conferences are faced with a difficult challenge: they are given roughly 20 minutes to deliver a speech that is both unforgettable and inspiring. This thesis will examine how to craft speeches that are both memorable and inspiring.

The goal of this thesis is to advance the use of recurrent neural network language models (RNNLMs) for large vocabulary continuous speech recognition (LVCSR). RNNLMs are currently state-of-the-art and shown to consistently reduce the word error rates (WERs) of LVCSR tasks when compared to other language models. In this thesis we propose various advances to RNNLMs. The advances are: improved learning procedures for RNNLMs, enhancing the context, and adaptation of RNNLMs. We learned better parameters by a novel pre-training approach and enhanced the context using prosody and syntactic features. We present a pre-training method for RNNLMs, in which the output weights of a feed-forward neural network language model (NNLM) are shared with the RNNLM. This is accomplished by first fine-tuning the weights of the NNLM, which are then used to initialise the output weights of an RNNLM with the same number of hidden units. To investigate the effectiveness of the proposed pre-training method, we have carried out text-based experiments on the Penn Treebank Wall Street Journal data, and ASR experiments on the TED lectures data. Across the experiments, we observe small but significant improvements in perplexity (PPL) and ASR WER. Next, we present unsupervised adaptation of RNNLMs. We adapted the RNNLMs to a target domain (topic or genre or television programme (show)) at test time using ASR transcripts from first pass recognition. We investigated two approaches to adapt the RNNLMs. In the first approach the forward propagating hidden activations are scaled - learning hidden unit contributions (LHUC). In the second approach we adapt all parameters of RNNLM.We evaluated the adapted RNNLMs by showing the WERs on multi genre broadcast speech data. We observe small (on an average 0.1% absolute) but significant improvements in WER compared to a strong unadapted RNNLM model. Finally, we present the context-enhancement of RNNLMs using prosody and syntactic features. The prosody features were computed from the acoustics of the context words and the syntactic features were from the surface form of the words in the context. We trained the RNNLMs with word duration, pause duration, final phone duration, syllable duration, syllable F0, part-of-speech tag and Combinatory Categorial Grammar (CCG) supertag features. The proposed context-enhanced RNNLMs were evaluated by reporting PPL and WER on two speech recognition tasks, Switchboard and TED lectures. We observed substantial improvements in PPL (5% to 15% relative) and small but significant improvements in WER (0.1% to 0.5% absolute).