2nd Paris Workshop

on Games, Decisions, and Language

June 13-15, 2024

University Paris-Panthéon-Assas


Abstracts


Despoina Alempaki (Warwick Buisiness School)
"Deceptive Communication: Direct Lies vs. Ignorance, Partial-Truth and Silence"
(based on joint work with Valeria Burdea and Daniel Read)
Abstract: In cases of conflict of interest, people can lie directly about payoff relevant private information, or they can evade the truth without lying directly. We analyse this situation theoretically and test the key predictions in an experimental sender-receiver setting. We find senders prefer to deceive through evasion rather than direct lying. This is because they do nοt want to deceive others, and they do nοt want to be seen as deceptive. We also find receivers are highly sensitive to the language used to deceive, and are more likely to act in the sender’s favour when the sender lies directly.
Working paper

Gerrit Bauch (University of Bielefeld, Center for Mathematical Economics)
"Effects of Noise on the Grammar of Languages"
Abstract: We study a signaling game of common interest in which a stochastic noise is perturbing the communication between an informed sender and an uninformed receiver. Despite this inhibiting factor, efficient communication is possible for any kind of noise and improves upon babbling unless the noisy channel is uninformative. Endowing a compositional message space with the Hamming distance, we explore the impact of a a well-known noise channel from information theory on the grammatical structure of efficient communication. Under noise, relabeling of cells cannot be arbitrary, but has to assign distant messages to the most distant states. The more noisy the channel, the less frequent messages are used that describe states closer to the pooling action. Efficient communication under noise can be learned through the forces of evolution, but not every equilibrium is stable.
Working paper

Andreas Blume (University of Arizona, Eller College of Management)
"Language Games: Correlation through Non-Understanding, Dialogue, Inarticulateness, and Misunderstanding"
Abstract: A language game is a finite complete-information game preceded by pre-play communication with explicit constraints on players’ ability to produce and understand messages and on their knowledge of each other’s constraints. Players communicate directly and publicly but may not understand or may misunderstand each other’s messages. The paper gives conditions under which it is possible to implement correlated equilibria outside the convex hull of the set of Nash equilibria through language games. These conditions can be satisfied in games with any numbers of players, including two. In the game of Chicken it is possible to induce the entire set of correlated equilibria via a language game. Working paper  

Valeria Burdea (Ludwig-Maximilian University Munich)
"I'm Sorry If You Are: Two-Sided Conflicts and a Theory of Relative Blame"
(based on joint work with Shereen J. Chaudhry)
Abstract: Apologies are powerful tools for reconciliation, so understanding and addressing barriers to apologizing can enhance long-term cooperation. Existing research on apologies leaves critical blind-spots by focusing on one-sided conflicts, where only one person is in a position to apologize: We find that most recalled unresolved conflicts are two-sided, such that each person is in a position to apologize. Combining insights from existing research, we propose a theory of relative blame that parsimoniously explains behavior in two-sided conflicts and makes novel predictions. The main assumption is that in such conflicts people care not only about what they are apologizing for (absolute blame) but also how much responsibility they are taking for the conflict (relative blame). Further, people expect onlookers to make pragmatic inferences about relative blame based on the number of apologies offered: Apologizing alone signals the apologizer deserves all of the blame, whereas a return apology splits the blame. Because this presents a coordination problem, we use a game theoretic framework to analyze how preferences over the distribution of relative blame are expected to impact apologizing behavior. We test the theory’s predictions in eight studies (N = 1,264), substantiating behavioral patterns unique to two-sided conflicts, including that people are less likely to apologize for the exact same offense when they see the conflict as two- versus one-sided, that the decision to apologize first is influenced by perceptions of others’ knowledge states, and that people treat apologizing first as a risky choice. We discuss additional hypotheses and extensions to the theory.
Working paper 

Daniele Condorelli (Warwick University, Department of Economics)
"Cheap Talking Algorithms"
(based on joint work with Massimiliano Furlan)
Abstract: We simulate behaviour of independent reinforcement learning algorithms playing the Crawford and Sobel (1982) game of strategic information transmission. We show that a sender and a receiver training together converge to strategies approximating the ex-ante optimal equilibrium of the game. Communication occurs to the largest extent predicted by Nash equilibrium. The conclusion is robust to alternative specifications of the learning hyperparameters and of the game. We discuss implications for theories of equilibrium selection in information transmission games, for work on emerging communication among algorithms in computer science, and for the economics of collusions in markets populated by artificially intelligent agents.
Working paper

Françoise Forges (Dauphine University)
"Finite Sender-Receiver Games"
Abstract: An informed player (with finitely many types) sends a message (from a finite set) to an uninformed player who chooses one of finitely many decisions. The players' utility only depends on the type and the decision (the message is cheap talk), but is otherwise arbitrary. What outcomes can be achieved? Various solution concepts are surveyed, from Nash equilibrium to communication equilibrium, including some of their refinements.

Luca Gasparri (CNRS, Laboratoire Savoirs, Textes, Languages, University of Lille)
"Natural Linguistic Conventions"
Abstract:

Nicolas Rodriguez Gonzalez (University of Arizona, Eller College of Management)
"The Geometry of Thought in Common-Interest Cheap-Talk Games "
Abstract: This paper explores a multi-sender common-interest communication model. Senders are language-constrained and the state space is multidimensional, which makes full information transmission infeasible. Efficient communication allows the receiver to process information through convex sets provided that preferences satisfy a generalized single-crossing property. Furthermore, these convex sets can be characterized by a finite set of hyperplanes. This mode of information processing aligns with cognitive psychology literature, as working with convex sets imposes fewer cognitive burdens compared to dealing with mappings or arbitrarily shaped regions. The paper demonstrates that language heterogeneity becomes necessary to achieve efficiency when the receiver is sufficiently responsive to information. Finally, it establishes that efficient communication through languages with natural properties is generally unattainable, underscoring the tension between the strategic aspects of language formation and natural language properties.

Olivier Gossner (CNRS, École Politechnique)
"Strategic Types: A Discrete Language for Higher-Order Interactive Decision Making"
Abstract:

Michael Greinecker (École Normale Supérieure-Paris Saclay, Centre for Economics)
"Causality and Correlation in Game Theory and Science"
Abstract:

Aviad Heifetz (Open University of Israel)
"What the Language Game of Game Theory Cannot Express"
Extended abstract

Roni Katzir (Tel Aviv University)
"Implicit questions, explicit answers, and how we use both to negotiate the common ground"
Abstract:

Tomas Koblizek (Institute of Philosophy, Czech Academy of Sciences, Prague) and Rene Levinsky (Faculty of Arts, Charles University, Prague)
"Lying as a Language Game? A Game Theory Perspective"
Abstract: The aim of our paper will be to answer the question whether lying can be understood as a language game in its own right, or whether lying is just a violation of rules of a language game. This question appears already in Wittgenstein (1953), and is further discussed by Searle (1979) and Jacquette (2004). In our paper, we will answer this question against the background of game theory and selected game experiments in which lying has been applied (Dato et al. 2019, Fischbacher and Föllmi-Heusi 2013, Gneezy et al. 2013, Gneezy et al. 2018, Khalmetski and Sliwka 2019). We will show that lying across these experiments is not a deviation from rules, but neither a language game: it is a conglomerate of diverse language games whose implicit rules change depending on the given game context. Based on this premise, (i) we will describe the main types of ‘lie-speech-games’ as they appear in various game experiments; (ii) we will ask to what extent the term ‘lie’ is univocal in game theory and to what extent it is ambiguous.

Frédéric Koessler (CNRS, HEC Paris)
"A Belief-Based Approach to Signaling"
(joint with Marie Laclau and Tristan Tomala)
Abstract: This paper provides a geometric characterization of the set of interim equilibrium payoffs in signaling games. It encompasses the general class of signaling games without imposing constraints on utility functions, action, or type spaces, except for assuming a finite type set. We apply a tractable belief-based approach similar to that employed in the literature on repeated games with incomplete information and cheap talk games. This approach allows us to avoid specifying the prior, the strategies of the sender and receiver, and the associated belief system. It relies on incentive-compatible splittings of beliefs, resulting in a constrained convexification of the graphs of the non-revealing payoff correspondences. Our characterization extends the existing equilibrium characterization established in sender-receiver cheap talk games, with and without assuming that the sender’s preferences are state independent. The characterization is illustrated in some classical signaling game applications and is used to derive the best equilibrium payoff of the sender when his preferences are type-independent.

Christoph Kuzmics (University of Graz)
"Fool me once, ..."
Abstract: We study repeated interactions (with perfect monitoring) between an (in every stage differently) informed sender and a receiver who takes actions and in which there is a conflict of interest between sender and receiver. The usual folk theorem applies. In particular, there is a subgame perfect equilibrium in which the sender always reveals the truth and the receiver does what they like best given the truth (the receiver's optimal outcome). One way to obtain this outcome as a subgame perfect equilibrium is based on a grim trigger punishment strategy, in which players play the (inefficient) equilibrium of the stage game (in which the sender lies and the receiver does their best given that). In other words, this is the strategy implied in the proverb "Fool me once, shame on you, fool me twice (or more often one would presume), shame on me." This subgame perfect equilibrium is, however, not renegotiation-proof in the sense of Farrel and Maskin (1989): the punishment is bad for both players, and they have an incentive to renegotiate. In the receiver-optimal renegotiation-proof strategy of this repeated game, the receiver obtains a strictly lower payoff than in the aforementioned subgame perfect equilibrium. This receiver-optimal renegotiation-proof strategy has the following properties: The sender is always truthful and the receiver mixes (or alternates) between using this information (to enhance their own payoff) and choosing an action that the sender prefers. Upon any deviation by the sender (i.e. when the sender is caught lying), the sender is requested to be fully honest and the receiver will fully use this information to their advantage for some pre-described time. Upon repeated deviations, the punishment strategy is restarted. This renegotiation-proof strategy seems a better (and more plausible) way to teach kids to be honest than the grim trigger strategy. 

Marco LiCalzi (Università Ca' Foscari Venezia)
"Vocabulary Aggregation"
Abstract:

Salvador Mascarenhas (École Normale Supérieure, Institut Jean Nicod)
"Questions in Language and Thought"
Abstract:

Marieke Pahlke (Corvinus University, Budapest)
"Dynamic Consistency and Ambiguous Communication"
Abstract: In most models of ambiguous communication, a Sender can only benefit from ambiguous communication if the Receiver behaves dynamically inconsistently. A dynamically inconsistent Receiver might not follow his ex-ante optimal plan after observing an ambiguous message. This paper proposes a novel approach to analyze ambiguous communication by studying dynamically consistent behavior in games with ambiguous strategies. We show that gains from ambiguous communication can be maintained even if players behave dynamically consistently. To achieve this, we define rectangularity, a condition on beliefs that ensures dynamically consistent behavior, for settings where ambiguity arises due to ambiguous strategies. Then, we analyze a Perfect Bayesian Equilibrium in an ambiguous persuasion setting. In this equilibrium, ambiguous communication outperforms standard Bayesian communication even if the Receiver behaves dynamically consistently. Finally, we extend our analysis to settings with ambiguous communication in cheap talk and mechanism design.
Working paper

Philip J. Reny (University of Chicago)
"Natural Language Equilibrium: Signaling Games"
Abstract:

Ariel Rubinstein (Tel-Aviv University and New York University)
"Convexity, Differentiability, and Language: A Comment on the Culture of Economic Theory"
Abstract:

Gerhard Schaden (University of Lille, Laboratoire Savoirs, Textes, Langages)
"Explaining the Euphemism Treadmill"
Abstract: There is a rich literature on the grammar of social meaning in language. However, there has not been much research into the questions of a) how linguistic expressions acquire social meaning in the first place; and b) how linguistic expressions with social meanings may change over time. Pinker named the "Euphemism Treadmill" as one instance of such a phenomenon. Unfortunately, his description is very short, and far from being a comprehensive. The aim of this talk is to try to delimit ranges of possible answers to these two questions, in a way that can integrate with the existing literature on the synchronic behavior of expressions with social meaning. Special attention will be paid to how speaker attitudes towards a) the meaning and reference of linguistic expressions; and b) other (types of) speakers might be reflected in these speaker's language use, and ultimately, grammar. I assume that agents have an attitude wrt denotata or intension of words in the interval [-1,1], where -1 is the most negative attitude, +1 the most positive attitude, and 0 perfect indifference, and that that for given intensions, there are several synonymous words available (w_0 .. w_n). Furthermore, every agent self-identifies as one of three types, namely {Hater, Lover, Indifferent} with respect to the intension, and signals according to this self-identified type. If there is consensus in the population of agents wrt to the attitude to the intension, there will be no further development. However, there will be some attitudes that will differ in the population (e.g., wrt Taylor Swift, Hamas, dogs, etc.). If there is some tendency for agents to not wanting to be mistaken to belong to another type, I will show that there will be pressure to develop a consensus for the synonyms, which will specialise for one of the three types - and thus, social meaning will have emerged. I will show that the Euphemism treadmill will come into place when social pressure exists to signal higher than one's actual attitude, and when agents have a (limited) capacity to do so.

Stephan Semirat (Université Grenoble Alpes)
"Indicative versus Imperative Meaning in Cheap-Talk Games: An Experiment"
Abstract:

David Spector (CNRS, Paris School of Economics)
"Comparing Alternative Game-Theoretic Models of Language Change: Replicator Dynamics, Fictitious Play, and Myopic Best-Response"
Abstract:

Bernhard von Stengel (London School of Economics)
"Large Language Models for Learning Strategies in Complex Games"

Nick Zangwill (University College London)
"Miscommunication in Evolutionary Language Game Theory"
Abstract:



LEMMA Université Panthéon-Assas Labex IJN