Humans make sense of language in context, bringing to bear their own understanding of the world including their model of their interlocutor's understanding of the world. In this course, we will explore risks that arise when we as humans bring this sense-making capacity to interactions with artificial interlocutors. In this class, we will investigate such questions as:
Course projects are expected to take the form of a term paper
analyzing some particular NLP product in terms of the concepts
developed through the quarter or proposing design strategies to
mitigate some of the risks we identified. Course projects can be
done individually or in pairs.
Prerequisites: Graduate coursework in compling/NLP or undergrad/graduate coursework in pragmatics
If you have already established accommodations with Disability
Resources for Students (DRS), please communicate your approved
accommodations to me at your earliest convenience so we can discuss
your needs in this course.
If you have not yet established services through DRS, but have a
temporary health condition or permanent disability that requires
accommodations (conditions include but not limited to; mental health,
attention-related, learning, vision, hearing, physical or health
impacts), you are welcome to contact DRS at 206-543-8924 or
uwdrs@uw.edu
or disability.uw.edu. DRS
offers resources and coordinates reasonable accommodations for
students with disabilities and/or temporary health conditions.
Reasonable accommodations are established through an interactive
process between you, your instructor(s) and DRS. It is the policy and
practice of the University of Washington to create inclusive and
accessible learning environments consistent with federal and state
law.
Washington state law requires that UW develop a policy for
accommodation of student absences or significant hardship due to
reasons of faith or conscience, or for organized religious
activities. The UW's policy, including more information about how to
request an accommodation, is available at Faculty Syllabus Guidelines
and Resources. Accommodations must be requested within the first two
weeks of this course using the Religious Accommodations Request form
available
at https://registrar.washington.edu/students/religious-accommodations-request/.
| Date | Topic | Reading | Due |
|---|
| 3/27 |
Introduction, organization, overview
|
No reading assumed for first day |
|
| 3/31 |
|
|
KWLA papers: K & W due 11pm |
| 4/3 |
Language as joint activity, intersubjectivity |
Clark 1996, Baldwin 1995, Reddy 1979, Weizenbaum 1976
- Reading questions:
- What does this say about how humans use language in communication?
- How do people react to computer-generated language?
- What are the crucial differences between human-human and human-machine communicative contexts?
- What sort of extra-linguistic information do people use and/or infer in human-machine or human-human interaction?
- What can we learn about the nature of dialogue as multi-turn conversations with changing salience over time/how that differs from one-turn conversations?
| |
| 4/10 |
Presupposition, Public commitments, Coherence relations |
- Readings:
- Bender & Lascarides 2019 (access via UW libraries, items 4-6 + 66-70 + 81 + 96)
- Hamblin 1970: Not available, so look at Walton 1991 instead ("Hamblin on the Standard Treatment of Fallacies")
- Lascarides & Asher 2009 (Agreement, Disputes and Commitments in Dialogue)
- Asher & Lascarides 2013 (Logics of Conversation; print book available at Suzzallo)
- Kim et al 2021 (Which Linguist Invented the Lightbulb? Presupposition Verification for Question-Answering)
- Kaplan 1978 (Indirect Responses to Loaded Questions)
- Levinson 2000 (Presumptive meanings, access e-book via UW libraries)
- de Marneffe et al 2019 (The CommitmentBank: Investigating projection in naturally occurring discourse)
- Reading questions:
- What are the structures involved -- what is a presupposition, what is an implicature, what is an entailment?
- How do people notice that the other party hasn't gotten or has misunderstood the implicature?
- How would a probabilistic model detect implicatures? / How could a probabilistic model control implicatures arising from its output?
- What would the desirable way for an artificial agent to reject faulty premises? And how and why?
- Are there any things that should not be presupposed at all?
- By artificial agents
- By humans in general
- By humans interacting with artificial agents
- Metrics: How are presupposition detection systems evaluated?
- How do public commitments fit with accountability (including by corporations)?
| |
| 4/17 |
Negotiating meaning (sociolinguistic perspectives) |
- Readings
- Erving Goffman (1974) Frame Analysis: an Essay on the Organization of Experience
- Sally McConnell-Ginet Words Matter
- Mische, Ann. 2003. Cross-Talk in Movements: Reconceiving the Culture-Network LinkLinks to an external site.. In Social Movements and Networks: Relational Approaches to Collective Action, edited by Mario Diani and Doug McAdam, 258–80.
- Myrendal, Jenny. 2019 Negotiating meanings online: Disagreements about word meaning in discussion forum communicationLinks to an external site.. Discourse Studies
- Rebecca Clift, 2001. Meaning in Interaction: The Case of Actually, Language, Vol. 77, No. 2, pp. 245-291 https://www.jstor.org/stable/3086775
- Hayashi et al (eds) 2013. Conversational Repair and Human Understanding. CUP
https://www.cambridge.org/core/books/conversational-repair-and-human-understanding/8698A66FEC0269B9BB6B96D052709AF0
- Steele 2019 Non-binary speech, race, and non-normative gender: Sociolinguistic style beyond the binary
- Moore 2006 ‘You tell all the stories’: Using narrative to explore hierarchy within a Community of Practice
- King 2016 On Negotiating Racial and Regional Identities: Vocalic Variation Among African Americans in Bakersfield, California
- Eberhardt & Downs 2015 "(r) You Saying Yes to the Dress?": Rhoticity on a Bridal Reality Television Show
- Chun 2011 Reading race beyond black and white
- Calder & King 2022 Whose gendered voices matter?: Race and gender in the articulation of /s/ in Bakersfield, California
- Austen 2020 Production and perception of the PIN-PEN merger
- Reading questions
- How do human interlocutors manage conversational repair?
- How do human communities coordinate on meanings even as the meanings are changing?
- Can conversation analysis developed for studying human hierarchies (of power) be applied to understand biases in artificial agent output?
- What is the fine line between what counts as a generalization v. bias? What is accepted presupposition vs. what is bias?
- How do people go about developing ways of communicating/different writing styles in different online communities? How would we extend this when we interact with artificial agents?
- What factors affect the differential perception of words when used to describe people of different social categories (e.g. gender)?
- What about the design of artificial voices wrt to sociolinguistic variables -- should artificial agents have a social address? What does it mean if they just get the "default" forms?
| |
| 4/24 |
Interface design |
- Readings
- Pearl (2016) Designing Voice User Interfaces
- Shevat (2017) Designing Bots
- Hall (2018) Conversational Design
- Deibel & Evanhoe (2021) Conversations with Things: UX Design for Chat and Voice
- Jokinen & McTear (2009) Spoken Dialogue Systems (Other works by McTear: https://www.waterstones.com/author/michael-mctear/901018 )
- Balentine (2007) It's Better to Be a Good Machine Than a Bad Person: Speech Recognition and Other Exotic User Interfaces at the Twilight of the Jetsonian Age
- Vlahos (2019) Talk to Me: How Voice Computing Will Transform the Way We Live, Work, and Think
- Schlangen (2005) Modelling dialogue: Challenges and Approaches Künstliche Intelligenz .bib: https://clp.ling.uni-potsdam.de/bibliography/Schlangen-2005-2/
- Bonomo (2023) An Ethical AI Never Says "I"
- Reading questions
- What considerations are given to social address in bot design?
- What are the factors we would evaluate voice or chat interfaces on?
- When is a dialogue system useful vs a hinderance?
- Can we imagine designing NL tools in a way that is not necessarily conversational? What would that look like?
- Could there be different design conventions for task- or intent-based chatbots vs. open-ended LLM based chatbots?
- What user data is appropriate to collect with what levels of transparency and how to build in consent?
- Data from one user to facilitate interaction with them v. data to be used across users
- Who are the users that the designers are imagining? Are they considering children, etc?
- How do these authors design user studies?
| Term paper proposals due |
| 5/1 |
Value Sensitive Design |
- Readings (further options on https://vsdesign.org/publications/)
- Hendry, D. G., Friedman, B., & Ballard, S. (2021). Value sensitive design as a formative framework. Ethics and Information Technology, 23(1), 39-44.
- Hendry, D. G. (2020). Designing Tech Policy: Instructional Case Studies for Technologists and Policy Makers. Tech Policy Lab, University of Washington.
- Young, M., Magassa, L., & Friedman, B. (2019). Toward inclusive tech policy design: a method for underrepresented voices to strengthen tech policy documents. Ethics and Information Technology, 21(2), 89-103.
- Logler, N., Yoo, D., & Friedman, B. (2018). Metaphor Cards: A How-to-Guide for Making and Using a Generative Metaphorical Design Toolkit. Proceedings of the 2018 Designing Interactive Systems Conference, 1373-1386.
- Yoo, D. (2018). Stakeholder Tokens: a constructive method for value sensitive design stakeholder analysis. Ethics and Information Technology.
- Friedman, B., Hendry, D. G., & Borning, A. (2017). A Survey of Value Sensitive Design Methods. Foundations and Trends in Human-Computer Interaction, 11(2), 63-125.
- Magassa, L., Young, M., & Friedman, B. (2017). Diverse Voices: A how-to guide for creating more inclusive tech policy documents. Tech Policy Lab, University of Washington.
- Yoo, D., Huldtgren, A., Woelfer, J. P., Hendry, D. G., & Friedman, B. (2013). A value sensitive action-reflection model: evolving a co-design space with stakeholder and designer prompts. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 419-428.
- Friedman, B., & Hendry, D. (2012). The envisioning cards: a toolkit for catalyzing humanistic and technical imaginations. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 1145-1148.
- Woelfer, J. P., & Hendry, D. G. (2011). Designing ubiquitous information systems for a community of homeless young people: precaution and a way forward. Personal and Ubiquitous Computing, 15(6), 565-573.
- Freier, N. G. (2007). Children distinguish conventional from moral violations in interactions with a personified agent. CHI '07 Extended Abstracts on Human Factors in Computing Systems, 2195-2200.
- Nathan, L. P., Klasnja, P. V., & Friedman, B. (2007). Value scenarios: a technique for envisioning systemic effects of new technologies. CHI '07 Extended Abstracts on Human Factors in Computing Systems, 2585-2590.
- Friedman, B. (1996). Value-sensitive design. ACM Interactions, 3(6), 16-23.
- Friedman, B. (1995). "It's the computer's fault": reasoning about computers as moral agents. Conference Companion on Human Factors in Computing Systems, 226-227.
- Reading questions, TBD
- What is the definition of value sensitive design?
- How can we apply value sensitive design techniques in the design or evaluation of chatbots/voice assistants/other systems that interact using NL?
- What's in common and what is different across the methods of value sensitive design?
- How can organizations making LLM chatbots take into account the values of diverse users and the stated values of the company while maintaining broad appeal?
- Tension between supporting user values and the goal of being "politically neutral"
- How do you elicit participants' values, without steering or influencing them?
- How do policy and tech design influence each other?
- Given recent tech developments, what is missing from current tech policy?
- How do we make the business case for bringing value sensitive design into designing chatbots?
- What does value sensitive design have to say about how to run user studies?
- How do these studies manage participant privacy?
| |
| 5/8 |
Term project peer discussions |
| Term paper outline due (5/9) |
| 5/15 |
Black Mirror Writer's Room |
|
|
| 5/22 |
Regulatory proposals |
- Readings
- Reading questions, TBD
- What systems are considered "AI systems" for the scope of these laws?
- What are the risks (relevant to interactive text systems) that the legislation appears to be targeting?
- Are they looking at risks to individuals, communities, civilization, and over what kinds of time scales?
- According to these proposals, who has accountability for the output of natural language generation systems?
- What are the legal consequences for violations of these laws?
- Do the regulations apply to: system builders, system users, media coverage, other?
- How general are the regulations? Do they call out specific use cases/disciplines of "AI"?
- For the regulations not yet in effect, what will the impact be on existing already deployed products?
- Do the regulations account for factuality problems with text generating machines and how to they address this?
- Are the proposals additional laws or changes to existing ones to account for "AI"? (Ex: slander, identity theft, medical malpractice...)
- To what extent do these proposals address data theft?
- How does copyright apply to synthetic media based on others' art?
| Term paper draft due |
| 5/29 |
No class [Memorial Day holiday] |
| |
| 5/30 |
|
| KWLA papers due
Comments on partner's paper draft due |
| 6/8 |
|
| Term papers due |