Questioning language machines

In post-war France, Canguilhem addressed the organism–machine analogy to contest the rise of cybernetics’ mechanist paradigm in science and philosophy. He criticised the analogy because ‘life tolerates monstrosities’, whereas ‘there is no machine monster […] no mechanical pathology’—meaning that cybernetics could never express the abnormality of life (Canguilhem 1952/2022: 90). It appears as if Canguilhem defended the plasticity of the Organism, its capacity of exceeding the norm against the rigidity of the Machine. However, his argument was more profound. Revisiting the controversy between Franz Borkenau (1932) and Henryk Grossmann (1935), Canguilhem traced the rise of the mechanist paradigm to its disciplinary and economic roots, that is the exploitation of labour power and of life in general. Canguilhem (1952/2022: 84) in fact argued that: ‘The theoretical mechanisation of life and the technical utilization of the animal are inseparable.’ Today, a similar observation could be made regarding the language–machine analogy, as AI starts dominating labour automation, cultural production, and even the humanities. In the age of AI, one could argue that the theoretical mechanisation of language and the technical exploitation of culture are deeply imbricated.

To address the disputed language–machine analogy, I suggest looking at the constitutive role of language in the sphere of labour. This requires reversing a trajectory of analysis that has become habitual in critical theory. Rather than simply recording how the machine alienates language, one should recognise that language has contributed to the machine design since its inception. As Hegel (1805, 1807) argued in the Jena lectures and in the Phenomenology of Spirit, it is the abstraction of human activity—labour as much as language—that creates the machine. Labour and language are machinic in their own right before the machine comes to alienate them, pointing thereby to a dimension that exceeds mechanisation. To deconstruct the final segment of the machine–organism–language genealogy, this concluding part moves from Hegel’s insight and briefly highlights other moments in philosophy and political economy in which language is discussed in its economic role.

Long before digital computers came to model language, language once happened to be conceptualised as a machine. In the circles of Soviet avant-garde art, formalists Viktor Shklovsky and Boris Tomashevsky framed the literary text as a mechanical device (Steiner 1984). Their poetics must be read against the backdrop of Soviet drives toward industrial automation and labour politics: in this context, it would not be difficult to envision a text as a productive machine, as a structure of visible and invisible relations aiming at causing an effect upon a collective audience. One could also easily define the literary text as a machine that emerges from the division of labour and language. A decade later, the US linguist Leonard Bloomfield (1933: 24) argued that ‘the division of labour, and, with it, the whole working of human society, is due to language’, hinting that language is part of the inner constitution of labour. Language works, language produces, and the product of this work, whether material or immaterial artefacts, shares the same fabric of language. Texts and books are alive, they respond to a large network of social interactions, to the vast social abstraction of language. In short, before digital computers imposed their machine language, and well before AI took over with its language machines, Soviet formalists had already envisioned the machine of language. They conceived linguistic artefacts as elements of a larger productive mechanism, much as one could view LLMs today, but with a crucial difference. Their poetics was not productivist, but rather alienist, seeking through art the effect of ostranenie, or estrangement (Shklovsky 1917/1991).

In the circles of French structuralism, the main successor of Russian formalism, the analysis of language, on the other hand, began gravitating around the sphere of circulation rather than production. Saussure (1916/2011: 79) famously wrote that: ‘[In linguistics] as in political economy we are confronted with the notion of value; both sciences are concerned with a system for equating things of different orders—labor and wages in one and a signified and signifier in the other.’ Saussure’s analogy between linguistic sign and economic value captivated numerous structuralists and Marxists (Barthes 1964, Lefebvre 1966, Goux 1973; and recently Tomšič 2015, Khatib 2017, Ricci 2024). Marx (1867) himself spoke of a Warensprache, a language that commodities would speak to one another, and this image persuaded many to recognise language more and more often in the sphere of circulation, within the value form, rather than in production, within the labour form, although Marx himself warned:To compare money with language is not less erroneous. Language does not transform ideas, so that the peculiarity of ideas is dissolved and their social character runs alongside them as a separate entity, like prices alongside commodities. Ideas do not exist separately from language. Ideas which have first to be translated out of their mother tongue into a foreign language in order to circulate, in order to become exchangeable, offer a somewhat better analogy; but the analogy then lies not in language, but in the foreignness of language. (Marx 1858/1973: 162)

In the years of structuralism and Foucault’s attempts at a comparative epistemology of labour and language, the Italian philosopher Ferruccio Rossi-Landi (1968, 1977) provided a remarkable synthesis of these two notions. Rossi-Landi (1977: 139) criticised Saussure as ‘he does not appear to possess a theory of linguistic work, the most viable foundation for any theory of linguistic value.’ Moving from Hegel’s anthropogenetic characters of labour and language, Rossi-Landi described a homology between the production of material and linguistic artefacts. According to Rossi-Landi, the internal design of artefacts such as machines can be compared to the internal structure of linguistic artefacts such as verbal expressions and written documents. There exists a ‘syllogism of action’ (Hegel 1816/2010: 732) that innervates both material and linguistic production. For this reason, Rossi-Landi also criticised Bloomfield who gave priority to language over labour when he declared that division of labour is possible thanks to language. Rossi-Landi (1977: 71, note 23) suggested that ‘language in its turn, and with it the whole working of human society, is due to the division of labor’ (Caffoni 2026). Ultimately, Rossi-Landi (1977: 102) recognised the computer as the culmination of the mechanisation of labour and language.

Rossi-Landi’s synthesis remains instructive: it helps in thinking philosophy of language and political economy within a common analytic framework, and it opens a path for understanding how the automation of language—most visibly in LLMs—reconfigures both the production of meaning and the distribution of economic value. A few decades after Rossi-Landi, other Italian economists such as Christian Marazzi (1994) advocated for a linguistic turn in political economy to grasp the transformations of social relations and conflicts in post-Fordism. Marazzi proposed to recognise the role of language as living labour and antagonistic dimension before its capture by apparatuses such as communication networks and computers. Marazzi (1994/2011: 35) insisted on fashioning the Turing machine as a ‘linguistic machine’ and a new kind of assembly line centring labour management around language and computational logic. Marazzi, and other Italian philosophers from the tradition of Autonomist Marxism such as Paolo Virno (1990/1996), stressed in particular the economic role of collective knowledge, understood as general intellect and mass intellectuality (Pasquinelli 2019). Their emphasis on the collective and political dimension of knowledge can be seen as a prophecy of the global AI that came, a few decades later, to measure, mathematise and mechanise the free intellects of post-Fordism. Yet it should be noted that, in advocating for a linguistic turn in political economy and debates about labour during the 1990s, nobody could have expected the degree of linguistic automation that is taking place today with LLMs. What these positions, and the sociology of labour more generally, lacked was a theory of automation.