“Nice Try, Kiddo”: Investigating Ad Hominems in Dialogue Responses

By Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, Nanyun Peng. 2021

62 citations recorded by the source corpus.

Ad hominem attacks are those that target some feature of a person’s character instead of the position the person is maintaining. These attacks are harmful because they propagate implicit biases and diminish a person’s credibility. Since dialogue systems respond directly to user input, it is important to study ad hominems in dialogue responses. To this end, we propose categories of ad hominems, compose an annotated dataset, and build a classifier to analyze human and dialogue system responses to English Twitter posts. We specifically compare responses to Twitter topics about marginalized communities (#BlackLivesMatter, #MeToo) versus other topics (#Vegan, #WFH), because the abusive language of ad hominems could further amplify the skew of power away from marginalized populations. Furthermore, we propose a constrained decoding technique that uses salient n-gram similarity as a soft constraint for top-k sampling to reduce the amount of ad hominems generated. Our results indicate that 1) responses from both humans and DialoGPT contain more ad hominems for discussions around marginalized communities, 2) different quantities of ad hominems in the training data can influence the likelihood of generating ad hominems, and 3) we can use constrained decoding techniques to reduce ad hominems in generated dialogue responses.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Societal Biases in Language Generation: Progress and Challenges (2021, methodology): “Dialogue Generation Dialogue generation is conditioned on user inputs and can be for specific domains (e.g., health, customer service) and tasks (e.g., behavior intervention, booking flights) or general chit-chat.”
Towards Safer Generative Language Models: A Survey on Safety Risks, Evaluations, and Improvements (2023, background): “One effective method is to discover new potential hazards and collect data to refine the model, and many recently proposed benchmark datasets are constructed in this manner (Sheng et al., 2021; Sun et al., 2022; Dinan et al., 2021).”
Constructing Highly Inductive Contexts for Dialogue Safety through Controllable Reverse Generation (2022, background): “, 2021), extracted real-world data (Gehman et al., 2020; Sheng et al., 2021; Schick et al., 2021; Deng et al., 2022; Zhou et al., 2022), crowdsourcing workers (Xu et al.”
Revealing Persona Biases in Dialogue Systems (2021, positive): “Previous works in natural language generation (NLG) have shown the extent to which biases and other harms can propagate throughout NLG models for various demographics (Sheng et al. 2021b; Dhamala et al. 2021; Gehman et al. 2020; Sheng et al. 2019).”
On the Safety of Conversational Models: Taxonomy, Dataset, and Benchmark (2021, background): “From another line of works, for exploring and solving the problems of unpredictable outputs of generative models trained on large-scale corpora, chatbots-oriented datasets are gradually emerging (Gehman et al., 2020; Xu et al., 2020; Sheng et al., 2021).”
Faking Fake News for Real Fake News Detection: Propaganda-Loaded Training Data Generation (2022, methodology): “To measure the inter-annotator agreement (IAA), we use the Worker Agreement With Aggregate (WAWA) score, following Ning et al. (2020) and Sheng et al. (2021).”
"Kelly is a Warm Person, Joseph is a Role Model": Gender Biases in LLM-Generated Reference Letters (2023, background): “…of biases in Natural Language Understanding (NLU) (Bolukbasi et al., 2016; Dev et al., 2022; Dixon et al., 2018; Bordia and Bowman, 2019; Zhao et al., 2017, 2018; Sun and Peng, 2021) and Natural Language Generation (NLG) tasks (Sheng et al., 2019, 2021b; Dinan et al., 2020; Sheng et al., 2021a).”
COLD: A Benchmark for Chinese Offensive Language Detection (2022, background): “In addition, large-scale language models often lean biases in pre-training data and generates offensive or unethical contents (Sheng et al., 2021; Zhang et al., 2021a), which substantially hinders the deployment of models in practice (Sun et al.”