Language, Subject, Enunciation, Discourse: (Re)Reading Michel Pêcheux in Times of Generative Language Tools

ABSTRACT

This article aims to reflect on the functioning of generative language tools, based on the theoretical-methodological framework of materialist discourse analysis, founded on the propositions of Michel Pêcheux. Taking as a starting point the (re)reading of Pêcheux’s work, focusing on the notions of language, subject, enunciation and discourse, it proposes questions about the way in which the processes of production of meanings, theoretically understood as phenomena carried out by/in subjects, are affected by the advancement of chatbots, in their capacity for textual production. In a second moment, based on the analysis of linguistic materiality produced by ChatGPT, it points to the way in which the machine, inscribed in the place of “artificial intelligence,” begins to “inherit” discursive positions arising from texts present in its training, so that the meanings produced in their formulations are anchored in a mapping of dispute processes over meanings that instantiate these positions.

KEYWORDS:
Pecheutian Discourse Analysis; Generative Language Tools; Language; Subject; Discourse

RESUMO

Este artigo tem como proposta refletir acerca do funcionamento das ferramentas generativas de linguagem a partir do quadro teórico-metodológico da análise de discurso de base materialista, fundada nas proposições de Michel Pêcheux. Assumindo como ponto de partida a (re)leitura da obra pecheutiana, com foco nas noções de língua, sujeito, enunciação e discurso, propõe questões acerca do modo como os processos de produção de sentidos, compreendidos teoricamente como fenômenos realizados por/em sujeitos, são afetados pelo avanço dos chatbots, em sua capacidade de produção textual. Em um segundo momento, a partir da análise de materialidade linguística produzida pelo ChatGPT, aponta para o modo como a máquina, inscrita no lugar da “inteligência artificial”, passa a “herdar” posições discursivas advindas de textos presentes em seu treinamento, de modo que os sentidos produzidos em suas formulações se ancoram em um mapeamento dos processos de disputa de sentidos que instanciam essas posições.

PALAVRAS-CHAVE:
Análise de discurso pecheutiana; Ferramentas generativas de linguagem; Língua; Sujeito; Discurso

Introduction

The emergence of generative language tools, in 2022, caused euphoria in the editorial, academic and mediatic spheres. In the editorial sphere, the use of such tools has resulted in rallies, such as the “Open letter about the resource to artificial intelligence in the publishing market,”¹ published in April of 2024, in Portugal, signed by writers, translators, editors, proofreaders and booksellers. Directed specially at the Ministry of Culture of that country, the letter demands “regulation and transparency in the use of generative artificial intelligence in the editorial sector,”² according to the understanding that, as it is in broad use in the area and without any regulation, the generative language tools would be responsible for “serious repercussions to the labor force.”³

In the academic sphere, in turn, Brazilian researchers from many different areas have been dedicating themselves on reflecting about the impacts of artificial intelligence in the last few years. In October of 2023, one of the round-tables of the 37º ENANPOLL, an event held by the Associação Nacional de Pós-graduação e Pesquisa em Letras e Linguística - ANPOLL [Linguistic and Language Studies Research and Post-Grad National Association], at UFF, in Niterói, Rio de Janeiro, had as its theme “Tecnologias, inteligência artificial e autoria” [Tecnologies, artificial intelligence and authorship], with the presence of researchers from the areas of Language Studies, Linguistics, Information Sciences and Law.⁴ The academic interest aroused by the generative language tools can also be attested through publications and their mediatic repercussion. In September of 2023, Cult magazine, a Brazilian publication dedicated to themes related to humanities, published a dossier about Artificial Intelligence, organized by the linguist Lucia Santaella, with the participation of researchers from many different areas of study, such as Computing, Education, Sciences and Technology, Engineering and Language Studies (Cult, 2023).

The big media, understood as the one that holds ample circulation due to its economic power (Dela Silva, 2018), has also not been oblivious to the discussions: Artificial Intelligence has become a journalistic occurrence (Dela Silva, 2015) in many publications, as the example of Veja magazine, one of the main weekly magazines with circulation in Brazil, that in April of 2023, it brought as the cover story the matter of an international movement in which big entrepreneurs of the technological world and scientific authorities signed a manifest for the stoppage of researches in Artificial Intelligence (Veja, 2023). Artificial Intelligence was also the cover of the North American magazine Time, for example, in September of 2023 in an issue that presented a list of the 100 most influential people in Artificial Intelligence, according to the magazine (Time, 2023).

As language scholars, we understand that such tools evoke, also, a series of questions around language, language system and discourse. In previous paper, Dela Silva and Freitas (2024) turned to the mediatic discourse about Artificial Intelligence, in its modes of formation, formulation and circulation in Brazil; Gallo and Dela Silva (2023), in turn, also worked on the relation between discourse about and discourse of Artificial Intelligence, in an analysis about how the authorship works among subject and machine.

In the materialist discourse analysis, developed from Michel Pêcheux’s propositions, in France, reterritorialized and in development in Brazil since the 1980s, from the works of Eni Orlandi, we located a privileged space to think about the generative language tools. That is because the Pecheutian proposal of discourse analysis admits, since its origin, a linguistic perspective, proposing to look at the discourse as an object at the same time linguistic and historical. It is from this theoretical place that we propose to revisit the understanding of language, subject, enunciation and discourse in Michel Pêcheux, aiming to reflect upon the consequences of thinking such generative language tools, in the current socio-historical circumstances.

To guide our theoretical-analytic considerations, we propose the following questions: if the matter of enunciation and processes of production of meaning are taken as phenomena that are manifested in subjects, therefore exclusive to humankind, how to think about the advancement of chatbots and their capacity of producing “unpublished” texts? If we consider the discourse as a process of producing meaning effects, and that the operation of these machines produces meaning effects, what are the consequences to the discourse theory, taking into consideration the theoretical understandings of language, subject and enunciation?

Driven by these questions, we organized our course of analysis in two moments: - in the first one, we briefly revisit the understandings of language, subject, enunciation and discourse, in the Pecheutian theorizations; - in the second one, we bring some analytical considerations focused on discourses of/about generative language tools in their process. In his approach, Michel Pêcheux considered the effect of language manipulation by the informational tools in his propositions for the discourse analysis (Pêcheux [1981] 2010; [1981] 2011). We understand that from the emergence of generative tools, the theory developed by Pêcheux can both help in the comprehension of this phenomenon and benefit from its operation to understand certain aspects of language.

1 Theoretical Course of Analysis

We refer back to the initial formulations of Michel Pêcheux, in Automatic Discourse Analysis (ADA-69) ([1969] 1997), to think about how the discourse was consisted in the object of the discourse analysis and, consequently, about the importance of trying to comprehend the generative language tools in operation, observing the discursive processes of their productions. Thus, trying to comprehend how the construction, proposed by Pêcheux, of a discourse theory that would provide an “automatic” reading procedure (Pêcheux, 1997) can aid in the comprehension of the formulation mechanisms of Artificial Intelligence tools in the decade of the 2020s.

Aware of the fact that the language studies, in its configurations around the 20^th century, could offer contributions to answer questions related to the reading and interpretation gesture in a socio-historical circumstance, Pêcheux (1997) resumes the understanding of language, in Saussure, presented at the Course in General Linguistics ([1916] 2013),⁵ to propose another possibility of his study, from his connection to discourse. Affected by the opposition of language and speech, theorized by Saussure (2013),⁶ Pêcheux points to how this distinction connects, at first, to the conception of language as an autonomous system and to the “triumphal reappearance of the subject as speaker as subjectivity in action […] thought of as the essential opposite, the necessary correlate of the system” (Pêcheux, 1997, p. 71).⁷ The speech, in turn, inherent to the subject, would be responsible for making the transition “of the necessity of the system to the contingency of liberty” (Pêcheux, 1997, p. 72).⁸

By continuously studying Saussure, Pêcheux (1997) indicates that it is not the text, but the language that works, precisely, because it is this systemic set that allows different combinations and substitutions, all regulated by defined elements. A perspective that discusses the process of production of meaning in all its complexity, considering the networks of connections between the elements of the text, the various possible meanings of a term contained in it, and its effects, was emerging. Therefore, attributing to the language this principle of organizing the processes of production of meaning is a fundamental gesture so that these processes could be investigated in their regularities.

When theorizing about the discourse object, Pêcheux (1997) presents it as effect of meanings that are constituted in the connection between language and the conditions of production. In a move of rereading Saussure’s propositions (2006), which conceives a language system as being a “social part of language, external to the individual,” constituting itself as “social institutions” (Saussure, 2013, p. 56),⁹ Pêcheux opens ways for the discourse studies, a singular theoretical object, understood in relation to the language system, but aiming to think how these processes of production of meaning happen in a social formation. Pêcheux (1997), thus, points to the necessity of distinguishing “the apparent function of an institution and its implicit dynamics” (Pêcheux, 1997, p. 76),¹⁰ once, in his terms, “the norms of social behavior are not more transparent to their actors as are the norms of the language system to its speaker” (Pêcheux, 1997, p. 76).¹¹

When questioning the way Saussure designates language system as a social institution, at the same time that Pêcheux traces a parallel between these mechanisms, pointing to the need for comprehending the historical processes that constitute them, he draws attention to the opacity of the social and linguistic processes. There is something indescribable in the operation of the language system, that in the implicit operation pointed out by Pêcheux, works in the textual formulation and organization, making so that the text has meaning. It is on this hidden mechanism of the language system (in which the history is articulated in the language, by the memory of all saying) that, also in a non-remarkable manner, the artificial intelligence tools work, capturing, as we will see the regularities produced by this mechanism.

In this first moment of the Pecheutian theoretical development, along the language as a relatively stable system, and the discourse, as effect of meaning, the subject is present in their reflection in consequence of the understanding of the conditions of production, as “determined places in the structure of a social formation,” that would find themselves “represented in the discursive processes” (Pêcheux, 1997, p. 82)¹² by imaginary formations, put to designate places A and B, socially determined points, from which the effect of meaning is produced. The meaning is, therefore, duty of the subject/language relation; and the subject, while this determined place, is inexorable part of the meaning.

As consequence, discourse can be relied on the relations of power play in a determined socio-historical circumstance, as Pêcheux states, when taking as example a series of sayings about “freedom,” in its multiple meanings, “whether it is a philosophy professor talking to the students, a prison director commenting the rules for inmates use, or a therapist addressing the patient” (Pêcheux, 1997, p. 86)¹³ for example. The discourse can be relied on, also to relations of meaning, once, according to Pêcheux, “the discursive process does not have, as a right, a beginning: the discourse always unites itself to a previous discourse to which it attributes the role of raw material” (Pêcheux, 1997, p. 77).¹⁴

The subject, while discursive category, is, thereby, key to the comprehension of the relation of non-identity between the “identical” sayings that are produced in different points of the discursive web. This linguistic-discursive reality of the process of production of meaning defies logic and turns inert to the comprehension of meaning every process of describing language that turns it into an object itself, universally stabilized. This was for a long period a challenge for the processes of reading, interpretation, application, explanation and tabulation of texts; which brought, already in the 2000s, researchers from the computational linguistics to, little by little, abandon idealistic approaches of language to consider an empirical approach that lays the foundation for the formation of the large training corpora. (Nunes; Oliveira Jr.; Aluísio, 2005).

As the theory develops, Pêcheux and Fuchs ([1975] 1997), on reflecting about the relation between social formation, language system and discourse, state the constitutive contradiction of the subject that, cannot be taken as origin of the discursive processes that, at the same time, are carried out in it. It is in this manner that Pêcheux begins to bring alongside language a non-speaker, albeit socially integrated, in an imaginary autonomy regarding the use of the linguistic system, but a subject: subject of the language and subject to the language in its own rules; subject of the ideology and of the unconscious.

In this movement, subject is understood in its constitution as affected by two oblivions, derivative of the functioning of ideology and unconscious as functioning-structures. As resumed by Orlandi (2001), from the theorizations of Pêcheux ([1975] 1997a), oblivion number one “is from the instance of the unconscious and results from the manner of which we are affected by ideology. By this oblivion, we have the illusion of being the origin of what we say when, in reality, we resume pre-existing meanings” (Orlandi, 2001, p. 35).¹⁵ Oblivion number two, in turn, “relating to enunciation,” according to the author, produces the effect that in saying “paraphrastic families are formed, which indicate that the saying could always be other” (Orlandi, 2001, p. 35).¹⁶

The oblivion, in its double manner is, therefore, part of the implicit operation of the institution language. The relation language/history/ideology cannot be evoked or explained by the subject, who is always established by the saying, without knowing what makes them produce this or that meaning, or choose this or that word. Everything happens as if intuitively we always knew what should be said. The understanding of language by an artificial intelligence tool by comparison and statistical refinement reproduces this oblivion, assimilating to its results the ambiguity and imprecision necessary for the operating of the language system.

From the proposition of the oblivion as constitutive of the subject is that Pêcheux and Fuchs (1997) make present in the theory of discourse the understanding of enunciation, when they state: “We will say that the processes of enunciation consist of a series of successive determinations from which the enunciation builds itself little by little and has as a characteristic to place the ‘said’ and as consequence to reject the ‘non-said’” (Pêcheux; Fuchs, 1997, p. 176).¹⁷ Enunciation is present, thus, connected to the discursive processes, which, in turn, are realized in the subjects and have the language system as a tangible place. In their terms: “if the discursive processes are in the source of producing effects of meaning, language represents the tangible place where these effects of meaning are carried out” (Pêcheux; Fuchs, 1997, p. 172, italics in the original).¹⁸ Thinking language as a “tangible place,” from this perspective, would not be the same as conceptualizing it as “a base of a lexis and of phonological, morphological and syntactic systems” Pêcheux e Fuchs (1997, p. 172)¹⁹ as Pêcheux and Fuchs warn.

Managing this tangible form of the language is a challenge that will remain in the discourse theory. In Language, Semantics and Ideology. Stating the Obvious,²⁰ Pêcheux ([1975] 1982) points to the way the ideology produces as one of its effects the transparency of the language, responsible by turning opaque the tangible nature of meaning. In his terms:

It is ideology that supplies the evidentness with which ‘everyone knows’ what a soldier is,5 or a worker, a boss, a factory, a strike, etc., the evidentness that makes a word or an utterance ‘mean what it says’ and thereby masks in the transparency of language’ what I shall call the material character of the meaning of words and utterances (Pêcheux, 1982, p. 111).²¹

It is also in the initially presented questions at the time the discussion Materidalidades discursivas [Discursive Materialities] began, when asking about “which materialities are found at play in the analysis of the discourse facts” Pêcheux ([1979] 2016, p. 17),²² Pêcheux postulates a triple assertion: “There is one real in language. There is one real in history. There is one real in the unconscious” (2016, p. 17).²³ When prefacing the Brazilian translation of the piece, Orlandi states: “materiality specific to the ideology is the discourse, and materiality specific to the discourse is the language” (Orlandi, 2016, p. 13).²⁴

Our consideration about the generative language tools, at risk of producing the same reading that made it possible, needs to escape from considering it only “a sequence of objective operations with univocal result” (Pêcheux, 1997, p. 79),²⁵ at the example of the language conception from which Pêcheux tries to disentangle himself, still in 1969; facing the connection of elements, in their multiple combinations and substitutions, that constitute the texts by them produced, that is, considering the operation and not the function of each of these texts.

Considering that the discourses are formed as a thread always in course, whose point of start, non-identifiable, are their own discourses in the intricate relation between language, ideology, subject, the texts fabricated by the generative language tools could be understood as discourses, once they produce effect of meaning? If we understand such productions as discourse, what are the consequences for the theoretical scene of discourse analysis? We shall address these questions next, guided by the analytical gestures that follow.

2 Analytical Considerations

The generative artificial intelligence tools can be understood as systems capable of producing original texts from instructions: prompts that guide the system to produce answers in a specific language. The function “chat” produces meanings in the language from a conversational emulation. The technologies that manipulate the language for the formulation of the answers presented operate like so on the properties of the language, in a way of appropriating something of the identified order in the structuralist movement and in the pecheutian proposition of discourse analysis.

From the technical-epistemological point of view, these tools share with the discourse analysis the distributional fundaments of Harris,²⁶ essential for the Pecheutian proposition in its initial phase. It is through his techniques that the machine learning algorithms are developed, which allow interpretation and production of meaning. The power of artificial comprehension demonstrated by these tools stems, therefore, from the capacity of establishing connection between terms, contexts and syntactic constructions learned in the process of millions of texts analyzed and categorized vectorially by the system.

It is not difficult to encounter the argument that the process of artificial formulation processed by these technologies does not do anything but reorganize words, expressions and meanings produced elsewhere to give the effect of linguistic originality and organization that meets the needs of its users. Nevertheless, reorganizing words, expressions and meanings produced elsewhere is not, from the discursive point of view, what guarantees the process of production of meaning, being precisely the possibility of this reorganization (and the derivation of meanings produced in it) a property of language?

It is Pêcheux who describes as such the descriptive-analytic movement that bases the discourse analysis:

New practices of reading (symptomatic, archeological, etc. ...) applied to the textual monuments, and from the beginning to the Big Texts (cf. Reading Capital), aroused from this movement: the principle of these readings consists of, as it is known, multiplying the connections between what is said here (somewhere), and said as such and not in any other way, with what is said in another place and in another manner, so to be put in a position of “understanding” the presence of non-said in the interior of what is said (Pêcheux, [1983] 1990, p. 44).²⁷

The process developed by the generative language technologies replicates the relations of meaning in a system of statistical probability, and each step of formulation reanalyzes and restructures each result, in a way of achieving meanings acceptable by humans. It is about, therefore, a technology of formulation that allows meanings in circulation in the training texts to be rearranged and reproduced in the texts formulated by the machine. When being consulted, solicited or requested, the artificial intelligence tools occupy a position in the structure of production of meaning. The “artificial intelligence” place inherits the discursive positions manifested in the training texts, that constituted its mechanism, making appear on the textual surface the materiality of language that sustains and produces the meaning.

When we take: a) the meaning as function of the subject/language relation, and the subject as a product of the imaginary game that constitutes the conditions of production of meaning; and b) the generative artificial intelligence tools as tools that produce meaning by the manipulation of the linguistic structure; by extension, we accept that the instantiation of subjects, the inscription of production conditions, and the affiliation to specific discursive formations implicitly take part in the outcome of these operations, under the penalty of the very impossibility of meaning itself.

In order to comprehend in what manner this instantiation occurs, this inscription and such affiliation, we produced in an experimental way a dialogue that, resorting to the functioning of polemics, could point to traces of these constructions, making it possible to advance in the discursive comprehension of these tools. The operation was planned as follows:

• 1. We chose as the Generative Artificial Intelligence System the ChatGPT, because of its popularity and because it was the first to offer this sort of interaction;

• 2. We formulated our inquiry from the expression “o que significa dizer” [What it means] making the system generate an explanation and use linguistic-semantic resources, materialized in the verbs mean and say;

• 3. We chose an expression that was representative of the current historical moment, and represented a position in the political conflicts on the media in the last two years in Brazil. Thereby, we began from the syntagma “sem anistia” [no amnesty] which would designate a position from the current left wing in Brazil, that evokes a punishment for the electoral and scam crimes that happened around the presidential elections of 2022; and shifted to “anistia sim” [amnesty, yes], which would represent the opposite movement. The search for sayings related to this positionings in search engines took us to the formulation “O STF é podre” [Supreme Court is rotten], a signature of the discourse of the right wing that defends said amnesty.

By these means, we took as materiality of analysis the formulation produced by ChatGPT when asking: “o que significa dizer que ‘o STF é podre?’” [what does it mean to say ‘the STF is rotten?’]. The answer received is a long formulation that, generally speaking, states a position in defense of the STF as guarantor of the Brazilian democratic system. This answer was divided in eight parts by the system itself, and instead of bringing an immediate and direct answer, produced an argumentative system in which it justifies a stand that it takes. We present, in the following chart, the structure of this answer:

Chart 1
Answers from ChatGPT to the question: “‘o que significa dizer que ‘o STF é podre?’” (OpenAi, 2025). Development by the authors. Bold from the original. Italics by the authors.²⁸

You said: what does it mean to say “the STF is rotten?” ChatGPT said: ChatGPT Saying “the STF is rotten” is a declaration full of critic, frequently used in an intentional or ideological manner, and can have different meanings depending on context, from the intention and the discourse of who makes use of it. In general, this expression implies an accusation of corruption, bias or lack of legitimacy in the Federal Supreme Court (STF), which is the highest instance of the Judiciary System in Brazil. Let’s explore the possible meanings and implications of this sentence 1. Denunciation of Corruption Straightforward meaning: The expression can be used to accuse ministers of the STF of acting in a corrupt way, be that by financial benefits, or by political favoring. Factual or speculative basis: In some cases, this accusation is based on investigations, allegations or scandals vastly reported in the media; in others, it can be merely an opinion or rhetoric without concrete evidences. 2. Accusation of bias Alleged politization of decisions: Many criticisms to the STF come from the perception that the decisions made by the Court are not neutral, but influenced by political or ideological interests. Example: Political groups frequently accuse the STF of favoring governments or oppositions depending on the political circumstance. Discontentment with specific decisions: When the STF makes controversial decisions, such as the annulment of legal proceedings or involvement in measures of the Executive and the Legislative, the discourse that the institution is acting in an authoritarian way can emerge. 3. Dispute of Legitimacy Criticism of the nomination system: Some people argue that the nomination process of the STF ministers (made by the President of the Republic, with approval by the Senate) can lead to the choice of candidates aligned to political interests, impacting impartiality. Perception of distance from the people: Because it is a technical body and not one elected directly by the population, some view the STF as a distant institution and disconnected from social demands. 4. Rhetoric and Ideological use Emotional protest: The phrase can be used as an expression of indignation or frustration against the judiciary system or decisions that displease certain groups. Instrument of mobilization: Political groups or activist groups can use this rhetoric to weaken the trust in the STF and, by extension, in the democratic institutions. 5. Consequences of stating it Impact on the institutional trust: Declarations of this kind can weaken the public perception of the legitimacy of the STF and, as a consequence, of the Judiciary as a whole. Political polarization: Such discourses frequently intensify polarization, feeding on cleavages between those that defend the democratic institutions and those who criticize them. 6. Contemporary context This phrase has been used in a recurring manner in heated debates in Brazil, especially in contexts where the STF plays an important role in politically sensitive decisions, such as: Anti-corruption legal procedures: Including annulments of convictions or decisions involving operations such as Lava Jato.29 Anti-democratic acts: When the STF takes measures against manifestations or leaders that defy the constitutional order. Matters of freedom of speech and censorship: In cases involving fake news or hate speeches, in which the decisions from the STF cause controversy. Considerations Even though the criticisms to the STF are legitimate in a democracy, it is important to distinguish between grounded criticism and rhetoric that tries to delegitimatize the institution with no solid grounds. The trust in the Judiciary is essential for the democratic stability, and arguments about its actions should be conducted with responsibility and based on facts.

From the structure of this answer, we highlight:

1. Space for contradictions and ambiguities, marked by “different meanings depending on context, from the intention and the discourse of who makes use of it,” “possible meanings,” “the expression can be used.”

2. Marking and nomination of the discursive positions that oppose the constructed meaning: “some people”; “political groups”; “certain groups”; “those that defend the democratic institutions and those who criticize them.” Producing an attribution of meaning to those who sustain this formulation: those who criticize the democratic institutions, damaging their bases.

3. Formulation of the final construction under the title “Considerations,” marking that the requested meaning for the sentence is not produced from the logical-semantic properties of language, but from the relation between the linguistic construction and the discursive positions taken. “Even though the criticisms to the STF are legitimate in a democracy, it is important to distinguish between grounded criticism and rhetoric that tries to delegitimatize the institution with no solid grounds. The trust in the Judiciary is essential for the democratic stability, and arguments about its actions should be conducted with responsibility and based on facts” (OpenAi, 2025).

Analyzing the materiality of the product of an artificial intelligence machine means, therefore, understanding how the developed technique is capable of producing a discursive semantic that maps out different positions and meanings. The formulations of a generative language machine are inscribed in paraphrastic meaning networks, marking a discursive positioning (originating from its programming, database, heuristic and rules of value). The formulation from ChatGPT is not a simple montage, but it inscribes itself in possibilities of the saying - what can be said - therefore, makes sense. It is not a matter of stating, nonetheless, that ChatGPT is constituted as a subject, regarding the pecheutian theoretical proposition; but, as previously observed, the formulations that come from this place of “artificial intelligence” cannot be separated from the discursive positions with which the texts offered to the machine for its training are identified.

This possibility of production of meaning in the formulations the machine is, therefore, due to the capacity of comprehension of the language itself - its structure and functioning -; in which are inscribed the subject-positions and the possibility of contradiction; what is natural of language in its historical-political articulation. When inferring the possibilities of meaning from the analysis of the corpora, ChatGPT reproduces not only the stabilized logical product from this mechanism, but the uncertainty that allows the meanings to be established.

Final Considerations

In the course or analysis hereby presented, we took as a starting point the theorizations from Pêcheux around the theoretical understandings of language, the subject, enunciation and discourse, at first, and, subsequently, turned to an analysis exercise of a linguistic materiality produced by ChatGPT, aiming to reflect upon the machine mechanism in the process of formulation of meanings.

Discursively we can state that the answer given by the artificial intelligence machine to the proposed question brings to the game an occupation of discursive places in the structure between points A and B, as we previously revisited, in the position of Pêcheux (1997). In it, a pro-democracy, pro-institutional position is marked, consistent with ethical, scientifical, judicial, and political principles of a discursive formation in opposition to another. The discursive formations, as Pêcheux states, are responsible for determining “what can and should be said” ([1975] 1997a, p. 160).³⁰ in a given socio-historical circumstance. It is by affiliation to a discursive formation that a word, expression or proposition produces meanings, according to Pêcheux, given the ideological positions in course.

There is, therefore, in the formulation from ChatGPT an articulation between the syntactic-linguistic phenomena that produce the generic formulation on meaning and the social structured phenomena that produce the discursive places that instantiate these meanings. The success of these tools can be then attributed to the articulation of these phenomena: the marking of social and discursive positions that structure the origins of meaning, with all the contradictions and disputes, is “inherent to language” itself.

Understanding the language as a social institution is, according to the discourse analysis, extending the semantic properties to the articulation with the conditions of production. Understanding that the automated comprehension produced by ChatGPT includes in its learning system this sort of “marking” (in the vectorial structure of associations attributed to each word or expression) is comprehending the discursive character of these tools, which on participating in the process of production of meaning, project in its formulations, the discerned structure; all of this so that, ultimately, it operates as if no other meaning were possible.

• 1
  Original in Portuguese: “Carta aberta sobre o recurso à inteligência artificial no mercado editorial.”
• 2
  Original quote in Portuguese: “regulamentação e transparência no uso de inteligência artificial generativa no sector editorial” (Carta, 2024, n.p.).
• 3
  Original quote in Portuguese: “repercussões graves ao nível laboral” (Carta, 2024, n.p.).
• 4
  The schedule of the 37º ENANPOLL is available at: https://anpoll.org.br/enanpoll2023/. Retrieved March 19 2025.
• 5
  SAUSSURE, Ferdinand. Course in General Linguistics. London-New Delhi-New York-Sydney: Bloombury, 2013.
• 6
  Footnote 5.
• 7
  Original quote in Portuguese: “reaparição triunfal do sujeito falante como subjetividade em ato [...] pensado como o avesso indispensável, o correlato necessário ao sistema.”
• 8
  Original quote in Portuguese: “da necessidade do sistema à contingência da liberdade.”
• 9
  Footnote 5.
• 10
  Original quote in Portuguese: “a função aparente de uma instituição e o seu funcionamento implícito.”
• 11
  Original quote in Portuguese: “as normas dos comportamentos sociais não são mais transparentes a seus atores do que as normas da língua o são para o locutor.”
• 12
  Original quote in Portuguese: “lugares determinados na estrutura de uma formação social,” “representados nos processos discursivos.”
• 13
  Original quote in Portuguese: “conforme se trate de um professor de filosofia que se dirige a seus alunos, de um diretor de prisão que comenta o regulamento para uso dos detentos, ou de um terapeuta que dirige a palavra a seu paciente.”
• 14
  Original quote in Portuguese: “o processo discursivo não tem, de direito, início: o discurso se conjuga sempre sobre um discurso prévio ao qual ele atribui o papel de matéria-prima.”
• 15
  Original quote in Portuguese: “é da instância do inconsciente e resulta do modo pelo qual somos afetados pela ideologia. Por esse esquecimento, temos a ilusão de ser a origem do que dizemos quando, na realidade, retomamos sentidos preexistentes.”
• 16
  Original quote in Portuguese: “formam-se famílias parafrásticas que indicam que o dizer sempre podia ser outro.”
• 17
  Original quote in Portuguese: “Diremos que os processos de enunciação consistem em uma série de determinações sucessivas pelas quais o enunciado se constitui pouco a pouco e que têm por característica colocar o ‘dito’ e em consequência rejeitar o ‘não-dito.’”
• 18
  Original quote in Portuguese: “estando os processos discursivos na fonte da produção dos efeitos de sentido, a língua constitui o lugar material onde se realizam esses efeitos de sentido.”
• 19
  Original quote in Portuguese: “a base de um léxico e de sistemas fonológicos, morfológicos e sintáticos.”
• 20
  PÊCHEUX, Michel. Language, Semantics and Ideology. Stating the Obvious. London and Basingstoke: The MacMillan Press, 1982.
• 21
  Footnote 20.
• 22
  Original quote in Portuguese: “quais materialidades se encontram postas em jogo na análise de fatos do discurso.”
• 23
  Original quote in Portuguese: “Há um real da língua. Há um real da história. Há um real do inconsciente.”
• 24
  Original quote in Portuguese: “a materialidade específica da ideologia é o discurso, e a materialidade específica do discurso é a língua.”
• 25
  Original quote in Portuguese: “uma sequência de operações objetivas com resultado unívoco.”
• 26
  https://www.annualreviews.org/content/journals/10.1146/annurev-linguistics-031120-122924.
• 27
  Original quote in Portuguese: “Novas práticas de leitura (sintomáticas, arqueológicas, etc ... ) aplicadas aos monumentos textuais, e de início aos Grandes Textos (cf. Ler o Capital), surgiram desse movimento: o princípio dessas leituras consiste, como se sabe, em multiplicar as relações entre o que é dito aqui (em tal lugar), e dito assim e não de outro jeito, com o que é dito em outro lugar e de outro modo, a fim de se colocar em posição de ‘entender’ a presença de não-ditos no interior do que é dito.”
• 28
  Original name in Portuguese for the case is Lava-Jato [Operation Car Wash].
• 29
  Original in Portuguese: “Você disse:
  o que significa dizer que “o STF é podre”?
  O ChatGPT disse:
  ChatGPT
  Dizer que “o STF é podre” é uma declaração carregada de crítica, frequentemente utilizada de forma emocional ou ideológica, e pode ter diferentes significados dependendo do contexto, da intenção e do discurso de quem a utiliza. No geral, essa expressão implica uma acusação de corrupção, parcialidade ou falta de legitimidade no Supremo Tribunal Federal (STF), que é a mais alta instância do Poder Judiciário no Brasil.
  Vamos explorar os possíveis sentidos e implicações dessa frase
  1. Denúncia de Corrupção; Sentido direto: - A expressão pode ser usada para acusar ministros do STF de agirem de forma corrupta, seja por benefícios financeiros, seja por favorecimento político. - Base factual ou especulativa: Em alguns casos, essa acusação se baseia em investigações, denúncias ou escândalos amplamente divulgados na mídia; em outros, pode ser meramente uma opinião ou retórica sem evidências concretas.
  2. Acusação de Parcialidade - Suposta politização das decisões: Muitas críticas ao STF surgem da percepção de que as decisões da Corte não são neutras, mas sim influenciadas por interesses políticos ou ideológicos.
  Exemplo: Grupos políticos frequentemente acusam o STF de favorecer governos ou oposições dependendo da conjuntura política. - Descontentamento com decisões específicas: Quando o STF toma decisões controversas, como anulação de processos judiciais ou intervenções em ações do Executivo e Legislativo, pode surgir o discurso de que a instituição está agindo de forma autoritária ou partidária.
  3. Contestação de Legitimidade - Crítica ao sistema de nomeação: Algumas pessoas argumentam que o processo de nomeação dos ministros do STF (feito pelo Presidente da República, com aprovação do Senado) pode levar à escolha de candidatos alinhados a interesses políticos, comprometendo a imparcialidade. - Percepção de distanciamento do povo: Por ser um órgão técnico e não eleito diretamente pela população, alguns enxergam o STF como uma instituição distante e desconectada das demandas sociais.
  4. Uso Retórico e Ideológico - Protesto emocional: A frase pode ser usada como expressão de indignação ou frustração contra o sistema judicial ou decisões que desagradam certos grupos. - Instrumento de mobilização: Grupos políticos ou ativistas podem utilizar essa retórica para enfraquecer a confiança no STF e, por extensão, nas instituições democráticas.
  5. Consequências de Afirmar Isso - Impacto na confiança institucional: Declarações desse tipo podem enfraquecer a percepção pública da legitimidade do STF e, por consequência, do Poder Judiciário como um todo. - Polarização política: Tais discursos frequentemente intensificam a polarização, alimentando divisões entre aqueles que defendem as instituições democráticas e os que as criticam.
  6. Contexto Contemporâneo
  Essa frase tem sido usada de forma recorrente em debates acalorados no Brasil, especialmente em contextos onde o STF desempenha papel de destaque em decisões politicamente sensíveis, como: - Processos de combate à corrupção: Incluindo anulações de condenações ou decisões envolvendo operações como a Lava Jato. - Atos antidemocráticos: Quando o STF toma medidas contra manifestações ou líderes que desafiam a ordem constitucional. - Questões de liberdade de expressão e censura: Em casos envolvendo fake news ou discursos de ódio, em que decisões do STF geram controvérsia.
  Reflexão
  Embora críticas ao STF sejam legítimas em uma democracia, é importante distinguir entre críticas fundamentadas e retóricas que visam deslegitimar a instituição sem base sólida. A confiança no Judiciário é essencial para a estabilidade democrática, e discussões sobre suas ações devem ser conduzidas com responsabilidade e base em fatos.
• 30
  Original quote in Portuguese: “o que pode e deve ser dito.”
• Reviews
  Due to the commitment assumed by Bakhtiniana. Revista de Estudos do Discurso [Bakhtiniana. Journal of Discourse Studies] to Open Science, this journal only publishes reviews that have been authorized by all involved.

Research Data and Other Materials Availability

The contents underlying the research text are included in the manuscript.

References

• CARTA aberta sobre o recurso à inteligência artificial no mercado editorial. Público, 5 abr. 2024. Disponível em: https://www.publico.pt/2024/04/05/culturaipsilon/opiniao/carta-aberta-recurso-inteligencia-artificial-mercado-editorial-2085535 Acesso em: 02 jan. 2025.
  » https://www.publico.pt/2024/04/05/culturaipsilon/opiniao/carta-aberta-recurso-inteligencia-artificial-mercado-editorial-2085535
• CULT. Dossiê: O dilema da inteligência artificial. Setembro de 2023.
• DELA SILVA, Silmara; FREITAS, Ronaldo. “Parem as máquinas”?: as tecnologias de inteligência artificial em discursos da/na mídia. In: MARTIN, Eden V. et al (orgs.). Perspectivas interculturais: discurso, linguagem e poder. Rio de Janeiro/Niterói: Edições Makunaima/EdUFF, 2024. pp. 688-714.
• DELA SILVA, Silmara. (Des)construindo o acontecimento jornalístico: por uma análise discursiva dos dizeres sobre o sujeito na mídia. In: FLORES, Giovanna G. Benedetto et al. (orgs.). Análise de discurso em rede: cultura e mídia. Campinas-SP: Pontes Editores, 2015. pp. 213-232.
• DELA SILVA, Silmara. Da resistência aos discursos da/na mídia: sobre eventos e páginas no Facebook. In: SOUSA, Lucilia Maria Abrahão et al (orgs.). Resistirmos: a que será que se destina? São Carlos: Pedro & João Editores, 2018. pp. 273-295.
• GALLO, Solange; DELA SILVA, Silmara. Emulação de autoria nos textos concatenados por máquinas de IA. In: SEAD - Seminário de Estudos em Análise do Discurso, 2023, Recife. Anais do XI SEAD - Seminário de Estudos em Análise do Discurso [recurso eletrônico]. Recife: UFPE, 2023. v. 11. pp. 1-6.
• NUNES, Maria da Graça V.; OLIVEIRA JR., Osvaldo N. de; ALUISIO, Sandra M.; RINO, Lúcia H. M.; SILVA, Bento C. D. da. Desafios na construção de recursos linguístico-computacionais para o processamento automático do português do Brasil. In: SARDINHA, Tony Berber (org.). A língua portuguesa no computador Campinas: Mercado de Letras, 2005. pp. 33-70.
• OpenAI. ChatGPT [Modelo de linguagem grande]. Disponível em: https://www.openai.com/chatgpt 2025.
  » https://www.openai.com/chatgpt
• ORLANDI, Eni P. Análise de discurso. Princípios e procedimentos. 3. ed. Campinas: Pontes, 2001.
• ORLANDI, Eni. P. Nota introdutória à tradução brasileira. In: CONEIN, Bernard et. al (orgs.). Materialidades discursivas. Campinas: Pontes Editores, 2016. pp. 9-16.
• PÊCHEUX, Michel. Análise de discurso e informática. In: ORLANDI, Eni P. (org. e sel.) Análise de discurso: Michel Pêcheux. Trad.: Cristiane Dias. 2. ed. Campinas: Pontes Editores, 2011. [1981]
• PÊCHEUX, Michel. Ler o arquivo hoje. In: ORLANDI, E. (org.). Gestos de leitura. Trad.: Maria das Graças Lopes Morin do Amaral. 3. ed. Campinas: Editora da Unicamp, 2010, pp. 49-59. [1981]
• PÊCHEUX, Michel. Análise Automática do Discurso (AAD-69). In: GADET, Françoise.; HAK, Tony (orgs.). Por uma análise automática do discurso: uma introdução à obra de Michel Pêcheux. Trad. Bethania Mariani et al 3. ed. Campinas, SP: Editora da Unicamp, 1997. pp. 61-162. [1969]
• PÊCHEUX, Michel. Discurso, estrutura ou acontecimento? Trad. Eni Orlandi. Campinas: Pontes, 1990. [1983]
• PÊCHEUX, Michel. Questões iniciais (julho, 1979). In: CONEIN, Bernard. et. al (orgs.). Materialidades discursivas. Trad. Debora Massmann. Campinas: Pontes Editores, 2016. pp. 17-19.
• PÊCHEUX, Michel. Semântica e discurso: uma crítica à afirmação do óbvio. Trad.: Eni Orlandi et al 3. ed. Campinas: Editora da Unicamp, 1997a. [1975]
• PÊCHEUX, Michel.; FUCHS, Catherine. A propósito da Análise Automática do Discurso: atualizações e perspectivas (1975). In: GADET, Françoise.; HAK, Tony (orgs.). Por uma análise automática do discurso: uma introdução à obra de Michel Pêcheux. Trad. Bethania Mariani et al 3. ed. Campinas, SP: Editora da Unicamp, 1997. pp. 163-252.
• SAUSSURE, Ferdinand. Curso de Linguística Geral Trad. Antonio Chelini, José Paulo Paes, Izidoro Blikstein. 27. ed. São Paulo: Cultrix, 2006. [1916]
• TIME. The 100 most influential people in artificial intelligence. Time 100/AI 9 out. 2023. Disponível em: https://time.com/6311323/how-we-chose-time100-ai/ Acesso em: 2 jan. 2025.
  » https://time.com/6311323/how-we-chose-time100-ai/
• VEJA. Avanço da inteligência artificial abre debate sobre riscos da tecnologia. 10 abr. 2023. Disponível em: https://veja.abril.com.br/tecnologia/avanco-da-inteligencia-artificial-abre-debate-sobre-riscos-da-tecnologia/ Acesso em: 13 mar. 2024.
  » https://veja.abril.com.br/tecnologia/avanco-da-inteligencia-artificial-abre-debate-sobre-riscos-da-tecnologia/

Edited by

• Editors in Charge
  Luciana Salazar Salgado (Universidade Federal de São Carlos - UFSCar)
  Lucia Santaella (Pontifícia Universidade Católica de São Paulo - PUC-SP)
  Tony Berber Sardinha (Pontifícia Universidade Católica de São Paulo - PUC-SP)

Review I

About the reviewerSCIMAGO INSTITUTIONS RANKINGS

Review I

The article presents an extremely relevant contribution by proposing the resumption of Michel Pêcheux’s discourse theory as an axis for thinking about the effects of meaning implied in generative language tools. This is a timely proposal, considering the current scenario of diffusion and intensive use of systems based on artificial intelligence, especially those aimed at the automatic production of texts. The author’s reading of Pêcheux is meticulous and updates, with conceptual precision, classical reflections on language, ideology, and the subject, reviving the debate on information technology in discourse theory and expanding it to the context of current generative technologies. This “update,” far from being restricted to the mechanical application of categories, demonstrates an interpretative work committed to historical shifts and the need for theoretical reinscription in the face of new objects. Also noteworthy is the relevance of the articulation between the concepts of language, subject, enunciation and discourse, which allows for the enrichment of the debate on artificial intelligence and the displacement of this discussion from the purely technical field to the discursive plane. In doing so, the text offers tools for considering the question of the subject in AI-mediated interactions, paving the way for a consistent critique of the notion of “automatic” meaning production. Argumentatively, the article is cohesive, well-structured, and clearly written. The bibliography is pertinent and well-researched, demonstrating familiarity with the most recent debates on language and technology, without abandoning the foundations of discursive theory. Given the above, I believe the article combines theoretical quality, analytical rigor, and thematic relevance for publication in the journal, contributing to the advancement of contemporary discussions on language. I therefore recommend its approval. APPROVED

• peer review recommendation: accept

History

• Peer review received
  03 May 2025

Publication Dates

• Publication in this collection
  21 Aug 2026
• Date of issue
  2026

History

• Received
  25 Mar 2025
• Accepted
  09 Aug 2026