This chapter introduces the LENS model, arguing that linguistic content influences decision-making by eliciting emotions and shaping perceptions of personal and social norms, which then guide strategic choices. It reviews evidence that wording shapes affective reactions and norm perceptions, and that both emotions and norms causally shape behaviour in economic games and moral judgements. The chapter also surveys their interaction: how emotions can generate or reinforce norms and how norm violations evoke emotions. Finally, it motivates a quantitative agenda: measuring emotional and normative content of text (e.g., with large language models) to build language-based utility functions.

How do people make decisions, and what does ‘utility’ really capture? This chapter reviews the classical, utility-based foundations of decision theory (risk vs uncertainty, expected utility, maximin/minimax regret) and introduces a programme that aims to understand how utility is computed. It formalises economic and experimental economic decisions and games, emphasising a key methodological innovation: linguistic instructions are integral to the decision problem and shape utility, a point the book develops to quantify language-based utility using large language models. The chapter reviews systematic violations of payoff maximisation in one-shot and anonymous interactions: (i) bounded rationality; (ii) heuristics and biases (loss-aversion, endowment, status quo, present biases); and (iii) social preferences revealed in altruism, cooperation, trust, fairness and altruistic punishment, the equity–efficiency trade-off, and truth-telling. Together these regularities motivate behavioural economics and the search for utility functions that extend beyond payoff maximisation, to include social welfare, equity, and – critically – language.

This chapter introduces large language models (LLMs) through a primer on neural networks, backpropagation, and transformer architecture, and explains how scale, data, and alignment methods (SFT, RLHF) shape LLM behaviour. It surveys uses of LLMs as decision-making assistants in work, healthcare, policy, and information domains, highlighting productivity gains alongside risks around bias and misinformation. Turning to economic contexts, it compares LLM choices with human behaviour across risk, time, and social preferences, noting greater rationality but also differences (e.g., ‘optimistic’ altruism and framing susceptibility). The chapter argues that LLMs increase demand for language-based utility functions.

This concluding chapter argues that language is a first-order driver of economic behaviour and outlines where the research should go next. It extends the LENS framework beyond one-shot decisions to strategic settings shaped by beliefs, and outlines the co-evolution between language and behaviour. Large language models are proposed as virtual laboratories, while a quantitative utility approach must accommodate multidimensional, non-linear emotions and norms, and expand to visual cues (VENS). The chapter highlights applications – from policy design to norm-sensitive AI – alongside serious risks of manipulation, surveillance, and bias. It closes with a call for transparent, ethically governed models that explain and responsibly influence decisions.

This chapter surveys how moral content shapes behaviour through language. It considers moral foundations theory and morality-as-cooperation theory, outlining their dimensions, correlates, critiques, and refinements. It then reviews lexicon-based approaches to normative analysis, from moral foundations dictionaries to newer MAC-aligned resources, and discusses the limits of current tools in separating personal, injunctive, and descriptive norms. Building on these insights, the chapter proposes language-based utility functions (LiMoLNoS and LiMoLENS) that integrate normative and emotional valuations to explain choices in economic settings. Overall, it argues that morality is multidimensional and measurable in text, enabling models that connect linguistic framing to decision-making and behaviour.

The development of artificial intelligence (AI) and machine learning is leading to a revolution in the way we think about economic decisions. The Economics of Language explores how the use of generative AI and large language models (LLMs) can transform the way we think about economic behaviour. It introduces the LENS framework (Linguistic content triggers Emotions and suggests Norms, which shape Strategy choice) and presents empirical evidence that LLMs can predict human behaviour in economic games more accurately than traditional outcome-based models. It draws on years of research to provide a step-by-step development of the theory, combining accessible examples with formal modelling. Offering a roadmap for future research at the intersection of economics, psychology, and AI, this book equips readers with tools to quantify the role of language in decision-making and redefines how we think about utility, rationality, and human choice.

This chapter challenges economic consequentialism by testing behavioural equivalence across economically isomorphic decisions and documenting systematic violations. Evidence spans lying aversion in sender–receiver versus isomorphic Dictator Games; moral linguistic framings that reshape choices in Ultimatum, Prisoner’s Dilemma, Dictator, and Equity–Efficiency Trade-Off games; and the ‘dark side’ where moral labels strategically manipulate others and even increase corruption. Beyond social contexts, language also shifts intertemporal, risk, and ownership decisions (e.g., foreign-language effects). Together, the results imply that utilities depend not only on outcomes, probabilities, and timing but also on language that activates moral norms and beliefs, motivating a shift to language-based utility.

This chapter argues that language matters for economic decisions and that modern large language models (LLMs) can quantify this effect. After outlining the limits of lexicon-based tools, it examines BERT and MoralBERT, showing that generic sentiment scores struggle to predict human behaviour, while adding moral dimensions helps but the results remain imperfect. LLM-based chatbots (e.g., GPT-4) enable context-sensitive sentiment estimates that predict framing effects, particularly in Dictator Games. Building on this, the chapter formalises language-based utility functions that combine payoffs with sentiment or moral polarity and derives testable predictions. Evidence across Dictator, Equity–Efficiency, and Bribery games supports the approach, while highlighting caveats and aveThe chapter highlights applicationsnues for refinement.

This chapter surveys behavioural models across decisions and games that rationalise deviations from money maximisation and shows that, despite variety, they share a single foundation: economic consequentialism. It reviews bounded-rationality accounts (satisficing/aspiration dynamics, quantal response equilibria) and departures from expected utility (prospect theory) and from exponential discounting in intertemporal decisions (hyperbolic discounting), then introduces social-preference formulations (altruism à la Ledyard/Levine, inequity aversion a la Fehr–Schmidt and Bolton–Ockenfels, and models by Andreoni–Miller and Charness–Rabin). Finally, the chapter formalises the ‘economic representation’ of action profiles and defines economic consequentialism – utility as a function only of payoffs, probabilities, and timing for all parties – while noting limits that motivate alternative approaches.

This chapter introduces sentiment analysis as a bridge between language and decision-making, reviewing emotion taxonomies (valence/arousal, Ekman, Plutchik) and core tools (LIWC, ANEW, SentiWordNet) for measuring the sentiment tenor of text. It proposes LiMoLES, a utility function combining monetary payoffs with language-elicited sentiment, and tests it on framing effects in the extreme Dictator Game, showing that human-rated sentiment predicts altruism better than lexicons. The chapter extends LiMoLES to basic emotions, clarifies limits of lexicon intensity and context, and motivates a shift toward normative components (LiMoLNoS) when emotions alone cannot explain choices, ultimately setting up a broader LENS model of emotional and normative influences.