Speaking without intention: Austin’s Speech Act Theory in the Age of AI
Conversational AI is, nowadays, widely spread and investigated, shaping the way we interact with technology. Despite its extensive use, it is not clear if the technology we are experiencing at this time is actually able to communicate from a linguistic perspective. In this sense, it is important to study Conversational AI approaches in comparison with linguistic theories to find flaws and identify limits. These theories provide deep insights into language structure and usage, essential for creating more human-like conversational agents. Specifically, the aim here is to emphasise how the integration of the speech act theory can lead to a new frontier of conversational AI based on linguistics that, instead of representing language uses only on a statistical basis, can take into account aspects that make the conversation more human-like.
In the field of linguistics and of philosophy of language, the theory of speech acts proposed by J.L. Austin has sparked significant discourse and debate. Austin’s theory investigates the complexities of language use beyond mere communication of information, emphasizing the performative aspects of speech. However, as soon as we dig into artificial intelligence (AI) and its dialogical applications, a poignant question arises:
What should AI systems consider to truly perform speech acts with intentionality akin to human communication?
In this exploration, we dissect Austin’s theory of speech acts and disclose the profound implications it holds for the design and functionality of AI technologies.
J.L. Austin introduced the theory of speech acts in his work “How to Do Things with Words” (1962). According to this theory,
“to say something is to do something; or in which by saying or in saying something we are doing something.”
He claimed, therefore, that utterances not only convey information but also perform actions, thereby affecting the social reality in which they are uttered. More specifically, Austin categorized speech acts into three main types:
Locutionary Acts: The act of uttering words with a specific meaning and grammatical structure.
Illocutionary Acts: The intended effect or force behind the utterance, such as making a promise, issuing a command, or asking a question.
Perlocutionary Acts: The actual consequences or effects of the utterance on the listener, such as persuasion, convincing, or shocking.
Austin emphasized that illocutionary acts are crucial for understanding the performative nature of language. They require not only the utterance of words but also the speaker’s intention and the listener’s interpretation.
The ability to comprehend and execute illocutionary acts depends on different factors. Firstly, consciousness and subjective experiences are important to process awareness, beliefs, desires, and intentions, essential components of intentional behaviour. Precisely, while intention refers to the mere purpose of a speech act, the intentionality encapsulates the phenomenological explanation of ‘consciousness’, i.e., the conscious reason of a certain purpose (Lanigan, 1977). Secondly, emotional understanding is useful to profoundly comprehend meaning in communication (Ondé et al., 2023). Thirdly, human communication is closely linked to goal-directed behaviour. Humans formulate goals, make plans, and take actions to achieve those goals based on their intentions (Grice, 1975). Lastly, intentionality often relies on understanding the situational context, social cues, and implicit meaning (Grice,1975).
The theory of speech acts by J.L. Austin sheds light on the performative nature of language and the intricacies involved in communication beyond the mere transmission of information. The AI research landscape involving dialogical applications can benefit from the recovery of the theoretical aspects of linguistics and, more specifically, from the study of pragmatics inherent to the type of speech acts processed, their interpretation, and their nature. The importance of intentionality in dialogue underlies the principles governing communication. Indeed, intentionality stands as one of the seven criteria of textuality outlined by De Beaugrand and Dressler (1981), a prerequisite for the existence of a text, as listed below:
- cohesion, the set of mechanisms used to ensure the connection between textual parts (i.e., pronouns, anaphora, deitics);
- coherence, the semantic structure of a text and the logical and psychological structure of the concepts expressed;
- intentionality, the intention of those who produce a cohesive and coherent text;
- acceptability, a cohesive and coherent text produced with a certain intentionality must be accepted by the receiver against the background of a given social and cultural context;
- informativeness, the degree of predictability or probability that certain elements or information will appear in the text; the enunciative exchange is made possible by the fact that the sender and listener share a common ground (Clark, 1996) derived from previous portions of the utterance or from references to extralinguistic experience, according to which the sender will assume that the receiver can easily reconstruct the topic being discussed even if it is not explicitly formulated;
- situationality, the relevance (Grice, 1975) and appropriateness of a text within a given communicative situation;
- intertextuality, relates the text to other texts with which significant connections exist.
In conclusion, as we navigate the evolving landscape of AI, enriching research with linguistics-based principles presents both a challenge and an opportunity to enhance the interaction capabilities of Conversational AI systems in meaningful ways. Furthermore, applying linguistic insights can make such systems more explainable and adaptable to different situations and user needs. Thus, embracing linguistic theories not only deepens our understanding of language but also enables the creation of more effective human-computer interfaces.
References
(Austin, 1962) Austin, John Langshaw. How to do things with words. Harvard university press, 1962.
(Clark, 1996) Clark, Herbert H. Using language. Cambridge university press, 1996.
(De Beaugrande & Dressler, 1981) De Beaugrande, Robert-Alain, and Wolfgang U. Dressler. Introduction to text linguistics. Vol. 1. London: longman, 1981.
(Grice, 1975) Grice, Herbert P. “Logic and conversation.” Speech acts. Brill, 1975. 41–58.
(Lanigan, 1977) Lanigan, Richard L., Speech act phenomenology, The Hague, NL: Martinus Nijhoff, 1977.
(Ondé et al., 2023) Ondé, Daniel, et al. “The Role of Emotional Intelligence, Meta-Comprehension Knowledge and Oral Communication on Reading Self-Concept and Reading Comprehension.” Education Sciences 13.12 (2023): 1249.
