IntelliPaper
Abstract
This study focuses on the translation into English of an ethnographic text from Benin by ChatGPT, a new technology that utilises Artificial Intelligence (AI). The purpose of the study is to evaluate the translation in terms of syntax, semantics, and pragmatics to assess the strengths and weaknesses of this technology, presented as a formidable tool that is poised to replace translators and render them unemployed. The method used is qualitative. It interprets the text in the Source Language (SL) and evaluates the translation against criteria of fidelity to the meaning of the SL text, cohesion of the discourse in the Target Language, and respect for the cultural context. The data, manually extracted from the translated text, consists of errors and mistakes found in the translation. The data analysis is conducted following Andrew Chesterman's theory on the three aforementioned translation strategies. The results of the study reveal that contrary to the current propaganda, ChatGPT primarily engages in literal translation. It does not engage in oblique translation. Indeed, errors and mistakes of syntactic, semantic, and pragmatic nature are abundant. Procedures such as transposition, modulation, foreignization, domestication, adaptation, transediting, etc., are almost unknown to it. At the current stage, ChatGPT is a tool that contains a vast number of words and can effectively assist translators in their work. It is too early to envision a scenario where this technology would replace experienced translators. Current scientific research trends should incorporate ChatGPT.
Explore Digital Article Text
I. INTRODUCTION
This article focuses on the English translation of an ethnographic text titled THE ORIGINS OF THE BAATONU (Bariba) PEOPLE by ChatGPT, which stands for "Chat Generative Pre-trained Transformer." The article aims to evaluate the translation on syntactic, semantic, and pragmatic levels to assess the strengths and weaknesses of this new translation technology. The three aforementioned concepts have been the subject of numerous studies in linguistics. Indeed, in a book titled An Introduction to English Syntax, Miller (2002) explains why we study syntax.
We study syntax because it enables human beings to compose complex messages. Suppose a disgruntled worker utters the single word idiot! He or she might have muttered stupid, unfeeling, ignorant idiot, with four words combined into a phrase. The speaker might even have said That stupid, unfeeling, ignorant idiot is the new manager!, in which the phrase the new manager and the phrase that stupid, unfeeling, ignorant idiot are combined into a clause by means of is. (p. xii).
Miller also explains that syntax concerns how words are arranged to form word groups (phrases), how these word groups are arranged to form clauses, and how these clauses are arranged to form sentences. Moreover, he highlights the properties of clauses.
We can usefully begin by reviewing the major properties of clauses. [...] Each clause contains a finite verb; that is, a verb marked for tense. [...] Finite verbs are traditionally said to be marked for tense, person and number. [...] In each clause, the finite verb is accompanied by its complements and adjuncts. [...] iii Each clause is marked for aspect, and the aspect can be changed. iv Each clause has mood. (Ibid, pp.60-61).
These properties of syntax as described immediately immerse us in the linguistic perspective of the concept. However, in addition to the linguistic perspective of the concepts of syntax, semantics, and pragmatics, this article approaches them from a translation studies perspective. Indeed, it draws inspiration from the theory developed by Andrew Chesterman (2016) in a book titled Memes of Translation.
The classification I shall propose is a heuristic one. It seems to work in practice; it uses accessible terminology; it seems to differentiate enough, but does not get bogged down in "unportable" detail; and it is flexible and open-ended. It comprises three primary groups of strategy: mainly syntactic/grammatical (coded as G), mainly semantic (S) and mainly pragmatic (Pr). It acknowledges that these groups overlap to some extent; that pragmatic ones usually involve semantic and syntactic ones as well, etc.; and that strategies of different types often co-occur. (2002, p. 90)
The author concedes that there are overlaps between these three categories. Indeed, there are textual elements that straddle syntax and semantics, pragmatics and semantics, or syntax and pragmatics. Concrete examples drawn from the source language (SL) text and the target language (TL) text will be given to illustrate this reality. Chesterman (2002) acknowledges that his theory is an effort to structure several translation theories developed by various authors. He states: "The classification does, however, also represent an attempt to structure various proposals made by other scholars into an overall framework. Many of the strategies listed have obvious subtypes, but I shall not develop these in any detail here."
The method used in this paper is qualitative. It interprets the source language (SL) text and evaluates ChatGPT's translation against criteria of fidelity, discourse cohesion in the target language, and respect for cultural context. The data, manually extracted, consists of errors and mistakes observed in the translation. The study results reveal that ChatGPT essentially makes literal and word-for-word translations, which does not faithfully render the meaning of sentences in specific contexts, and does not recognize the grammatical functions of words (for example, subject, complement, etc.) in all cases. In several instances, ChatGPT does not perform transediting. Indeed, when a sentence is not well written in the source text, ChatGPT does not automatically correct it and translates it as is. When, for example, a comma is missing in a sentence, ChatGPT does not take this into account and translates the sentence as is, giving it a different meaning. The following pages provide more details on the results obtained.
Despite ChatGPT's numerous shortcomings in translation, it is useful to include it henceforth in scientific research tools.
II. LITERATURE REVIEW
This literature review aims, first, to explain the syntactic, semantic, and pragmatic strategies from a translation studies perspective, using Andrew Chesterman's book Memes of Translation: The Spread of ideas in translation theory (2016) as a theoretical framework. Given that ChatGPT is a software that is part of a long tradition of translation aid tools, a brief historical overview of translation softwares will be presented starting from the mid-1940s, which mark the beginning of machine translation or computer-assisted translation (CAT) tools. The third stage of this literature review is devoted to the functioning and capabilities of ChatGPT as a translation tool. An interview conducted with ChatGPT about its translation skills is also included.
2.1 Chesterman's Syntactic, Semantic, and Pragmatic Categories
About syntactic strategies, Chesterman explains that:
These may be thought of as involving purely syntactic changes of one kind or another. Larger changes obviously tend to involve smaller ones too. Syntactic strategies primarily manipulate form. The main ones are these: G1: Literal translation, G2: Loan, calque, G3: Transposition, G4: Unit shift, G5: Phrase structure change, G6: Clause structure change, G7: Sentence structure change, G8: Cohesion change, G9: Level shift, G10: Scheme change. (2016, p. 91).
This list contains strategies such as literal translation, calque, borrowing, and transposition that we already know since they are discussed in Jean Paul Vinay's and Jean Darbelnet's Comparative Stylistics of French and English (1958). We will endeavour to explain only the less familiar concepts such as Unit shift, Phrase structure change, Clause structure change, Sentence structure change, Cohesion change, etc., which are nevertheless addressed in J.C. Catford's A Linguistic Theory of Translation (1965). In explaining the concept of Unit Shift, Chesterman (op. cit., p.93) states:
This is a term from Catford (1965). The units are: morpheme, word, phrase, clause, sentence, paragraph. A unit shift occurs when a ST unit is translated as a different unit in the TT: this happens very frequently, of course, and subclassifications can be set up for unit shifts of different types. Examples: ST: Wir akzeptieren folgende Kreditkarten und ersuchen Sie, jene, mit der Sie Ihre Rechnung begleichen wollen, anzukreuzen.
TT: We accept the following credit cards. Please mark the one which you would like to have charged. [One sentence to two; German clause becomes English sentence.]
On page 93, Chesterman explains the notion of phrase structure change which includes a number of changes at the phrase level, notably 'number, definiteness and modification in the noun phrase, and person, tense and mood in the verb phrase.' A phrase from the source text can be translated by a phrase in the target text, but its internal structure changes. On page 94, the notion of Clause structure change appears. The changes concern the structure of the clause, which is composed of elements such as the subject, verb, object complement, adverb, voice (active or passive), finite or non-finite structures, transitive or intransitive form. The Sentence structure change induces changes at the sentence level, and the status of the clauses that make up the sentence. For example, a sentence composed of a single clause can become, in the translated text, a sentence with two clauses, namely a main clause and a subordinate clause. "ST: Schon der Name signalisiert ein sorgfaltig durchdachtes Qualitatsprogramm... TT: As its name suggests, this is a painstakingly devised quality program... [German main clause to English sub-clause plus main clause]." (Ibid, p.95).
Cohesion change affects intra-textual elements, for example ellipsis, substitution, pronominalization and repetition, etc. Examples: "ST: Diese Ausgabe... TT: In the present issue... [The German reference to "proximity" is made through a demonstrative; this changes to definite article + adjective.]" (Ibid) On page 96, the concept of Level shift appears. "In a level shift, the mode of expression of a particular item is shifted from one level to another." Chesterman explains that what some languages (like English) express in the form of interrogation, other languages (like German) express through word order, or morphology (the example of Finnish). "ST: Wir... ersuchen Sie,... anzukreuzen. TT: Please mark... [German expresses the sense of "polite request" here via lexis (a particular verb) plus syntax (main verb plus infinitive); English uses only the lexical item please.]" (Ibid).
The case of semantic strategies:
Under semantic strategies I group kinds of changes which mainly have to do with lexical semantics, but also include aspects of clause meaning such as emphasis. Semantic strategies manipulate nuances of meaning. Several of these strategies derive from Vinay and Darbelnet's concept of modulation. (Ibid, p.98)
The list of semantic strategies includes synonymy, antonymy, hyponymy, converse, abstraction change, distribution change, emphasis change, paraphrase, trope change, etc.
Buy and sell are an example of converse. Abstraction change involves moving from concrete to abstract or from abstract to concrete. Distribution change: This is a change in the distribution of the "same" semantic components over more items (expansion) or fewer items (compression). (Ibid, p.100) Paraphrase: This is a typical strategy for the translation of idioms, for instance, for which no corresponding idiomatic expression can be found in the TL. (Ibid, p.100) Trope change: This strategy applies to the translation of rhetorical tropes (i.e. figurative expressions).
The pragmatic strategies:
Pragmatic strategies tend to involve bigger changes from the ST, and typically incorporate syntactic and/or semantic changes as well. If syntactic strategies manipulate form, and semantic strategies manipulate meaning, pragmatic strategies can be said to manipulate the message itself. (Ibid, p.104).
Pragmatic strategies include 'cultural filtering, explicitness change, information change, interpersonal change, illocutionary change, coherence change, partial translation, transediting, visibility change. Chesterman (Ibid, p. 104) explains that "This strategy is also referred to as naturalization, domestication or adaptation." These concepts were developed by Vinay and Darbelnet and Venuti. "Explicitness change is either towards more explicitness (explicitation) or more implicitness (implicitation)". On page 106, the author explains that 'Information change' involves adding or removing information concerning the source text. In this case, the translator adds information not mentioned in the source text to make the translation more explicit. Conversely, they may remove information deemed irrelevant. 'Interpersonal change' implies a change in the relationship between the text, the author, and the reader. For example, the formal "vous" used in French as a sign of politeness does not exist in English. In this specific case, the translator might use, for example, the word "please" instead of the formal "vous". Furthermore, Illocutionary change, "for instance, changes the mood of the verb from indicative to imperative". (Ibid, p.107) Regarding coherence change, the author states that "Whereas the cohesion change strategy listed under G8 has to do with formal markers of textual cohesion, coherence changes have to do with the logical arrangement of information in the text, at the ideational level." (See e.g. Blum-Kulka 1986.) Finally, "Transediting is a term suggested by Stetting (1989) to designate the sometimes radical re-editing that translators have to do on badly written original texts." (Ibid, p.108).
2.2 Brief History of Translation Softwares
In a presentation on the history of machine translation titled "HUMAN TRANSLATION: POST-EDITING AND MACHINE TRANSLATION. MORE FEAR THAN HARM - A Historical Retrospective on the Failures of Computational Linguistics and the Future of Translation," Dr. Stéphanie Engola from the Pan-African University (Cameroon) distinguishes five major decades in the history of Machine Translation:
1945-1955: The first steps. 1955-1965: From enthusiasm to disappointment. 1965-1975: The quiet period. 1975-1985: The awakening. 1985-1995: Maturity.
Then, she recalls that Andrew Booth, a crystallographer, and Warren Weaver, vice president of the Rockefeller Foundation, are the pioneers of Machine Translation (MT). Dr. Engola recalls that in 1947, Booth made a false analogy between translation and cryptography and began his work with Richens. In 1951, research teams were formed on the West Coast (Los Angeles, Seattle) and the East Coast of the United States (Massachusetts Institute of Technology, MIT). This marked the beginning of research in machine translation (MT). Also in 1951, Yeoshua Bar-Hillel debated the impossibility of achieving purely automatic translation. He indicated the necessity of human intervention both upstream (pre-editing) and downstream (post-editing). The scientific reasons mentioned included: the difficulty of developing a universal grammar, the difficulty of transposing the grammar of one language into another, the specialisation of domains, the standardisation of languages, and artificial languages.
In 1952, the first conference on machine translation was organized at the Massachusetts Institute of Technology (MIT). In 1954, at Georgetown University, the first experiment of Russian-English machine translation was carried out on a computer in collaboration with IBM. This experiment fostered the growth of research in Machine Translation (MT) in the United States and in other countries such as Great Britain, Canada, Italy, and the USSR.
The period from 1955-1965 was characterized by the Cold War and massive subsidies to military bodies. MT nourished formal linguistics, which later became computational linguistics. The latter also inspired Artificial Intelligence (AI), which was born in 1956. At MIT, Ingve and Chomsky developed the formalisation of syntax. Working methods were diversified, but all teams started with direct translation (word-for-word), to which syntactic analysis was added. Three approaches were favored: the direct translation system, the interlingual system, and the transfer system. In 1959, Bar-Hillel published the report of a survey conducted on the evolution of MT research in the United States and Great Britain. The results were as follows: on the basis of the automatic treatment of ambiguity, he concluded that it was totally impossible to achieve a FAHQT (Fully Automatic High-Quality Translation). One can achieve mediocre automatic translation or realize a "high-quality translation by a machine-post-editor-partnership." In 1961, Mortimer Tauber published a book titled "Computers and Common Sense," in which he concludes: It is impossible to achieve an operational automatic system. It is impossible to formalize the language (relying on the impossibility of strict synonymy). After the installation of several systems, notably EURATOM, Ispra, IBM on Mark II, the automatic translations were of poor quality and always required human intervention. In 1964, Peter Toma developed SYSTRAN (SYStem TRANSLation) for English-Russian MT. This system was later used by NASA for the joint Apollo-Soyuz mission.
The decade from 1975 to 1985 was marked by two major events: Japan's entry onto the MT scene and the development of supranational organizations, notably the European Economic Commission (EEC) in 1975, the Pan American Health Organization in 1976, and the development of the Canadian TAUM-METEO program (English-French), which use machine translation. The decade from 1985 to 1995 was marked by the diversification of language combinations, the deepening of the transfer system, and the design of translation memories (Trados, Wordfast, MemoQ, Fluency). Despite several decades of technological and financial efforts, machine translation has not yielded the expected results.
2.3 Presentation of ChatGPT
In an unpublished work titled "An Assessment of a Historical and Cultural Text Translated by ChatGPT," Monnon (2024) reports that:
In the ever-evolving landscape of artificial intelligence (AI), ChatGPT emerges as a revolutionary conversational tool, bridging the gap between humans and machines with remarkable finesse... The journey of ChatGPT begins with the introduction of GPT-1 by OpenAI in 2018. GPT-1 was the first iteration of the GPT architecture, based on the transformer model, and was trained on a vast corpus of text from the internet. In February 2019, OpenAI unveiled GPT-2, a larger and more powerful version of the GPT architecture. OpenAI continued to push the boundaries of the GPT architecture with the release of GPT-3 in June 2020. (2024, p. 67)
This quotation explains that ChatGPT was trained using a vast corpus of text extracted from the Internet. This corpus provides it with an enormous amount of words and contexts of language usage. ChatGPT 4 already exists. Since it is an interactive application, ChatGPT 4 were asked questions about its translation skills. Its responses will be presented at a later stage.
III. METHODOLOGY
3.1 Problem Statement
The central issue of this research is the general outcry that ChatGPT is causing not only in academic circles but also in the professional world where some business leaders mistakenly believe that this new technology can handle the translation of their official documents. Furthermore, some university officials are questioning the necessity of maintaining translation training programmes due to the new translation technologies. Students also become hesitant about this field, which they believe offers less security because of stiff competition from translation softwares. Due to these doubts, it is time for translation professionals to take the floor and provide a professional opinion on the matter.
3.2 Method
This research adopts a qualitative approach by requesting ChatGPT to translate an ethnographic text and by evaluating it on syntactic, semantic, and pragmatic levels. The evaluation involves identifying translation mistakes and errors and subsequently revising the translation accordingly. A questionnaire was submitted to ChatGPT, which it answered. Below is an excerpt from the questions and answers:
Do you know how to translate? Yes, I can help with translation. Please provide the text you need to translate and specify the target language.
How did you learn to translate?
As an AI language model developed by OpenAI, I was trained on a diverse dataset containing text in multiple languages. This dataset includes books, articles, websites, and other text forms, allowing me to learn patterns and structures in different languages. My translation capabilities stem from this extensive training, which enables me to understand context, grammar, and vocabulary across various languages.
While I strive to provide accurate translations, my outputs are generated based on patterns in the data I was trained on and do not involve actual language comprehension as a human translator would have. Therefore, while I can handle many translation tasks effectively, consulting a professional translator is recommended for complex or nuanced translations.
Do you know how to translate idioms and metaphors across languages? Yes, translating idioms and metaphors across languages requires a nuanced approach because these expressions often rely on cultural references and specific language structures that do not translate directly.
Do you have any weak points in translation? While I strive to provide accurate translations, there are some inherent limitations and potential weak points in my translation capabilities: Context sensitivity, idioms and metaphors, cultural nuances, technical jargons, evolving language, emotional and tonal accuracy, ambiguity and multiple meanings, idiomatic and dialectal variations.
Some people fear that you take their jobs and render them unemployed. Are they right?
The concern about AI, including language models like me, taking jobs is understandable. AI can automate certain tasks, which might lead to job displacement in some areas. However, it's important to consider a few points: Task automation, job creation in fields like AI development, human-AI collaboration, skills evolution, economic and policy factors. While some job displacement is likely, the overall impact of AI on employment will depend on how society manages the transition and leverages the benefits of AI technologies.
3.3 Data
The data consists of numerous mistakes and errors that were manually identified and processed.
IV. RESULTS
Given the title of this article, the research results are presented on the aforementioned three levels.
Syntactic Errors
-
ChatGPT's translations require restructuring efforts from the reviewer. Additionally, the software sometimes omits words.
-
ChatGPT fails to recognize vernacular expressions (e.g., the expression: "O mon na O bourou bèè" was translated as "O my na O bourou bèè," which is meaningless).
-
ChatGPT's translations are often 'out of context' (e.g., the use of "these" instead of "they"; "it" instead of "he"; there is confusion in the use of pronouns).
-
Unfaithful translation to the source text.
-
ChatGPT does not understand some words used in specific contexts well and gives them a general meaning.
-
Poor translation of the French "On" – ChatGPT interprets 'on' as "we," while in some cases, 'on' is neutral.
Semantic Errors
-
ChatGPT makes many lexical semantics errors due to a poor understanding of the context of word usage (e.g., the word 'pays' is translated as 'country' instead of 'homeland.' Indeed, the word 'pays' is used in a specific context where it means 'homeland' (pays baatonu); furthermore, the word 'mot' is translated as 'word' instead of 'remarks' (the word 'mot' is used in a context where it means 'remarks'); the term 'vol exprès' is translated as 'express theft' instead of 'stealing on purpose'; 'groupements dominés' translated as 'subject groups' instead of 'dominated groups'; the word 'appellation' translated as 'appellation' instead of 'name').
-
ChatGPT does not use the technique of transposition.
-
The dynamic equivalence technique is unknown to ChatGPT.
10. ChatGPT sometimes produces incomplete sentences.
Pragmatic Errors
-
ChatGPT does not change the expressive mode (e.g., imperative instead of indicative when necessary; present instead of subjunctive when necessary).
-
Inability to avoid neologisms in the translated text. The software retains French neologisms that make no sense to the English reader.
-
ChatGPT does not understand the concept of modality (e.g., 'il serait malade' is translated as 'He would be sick' instead of 'He is said to be sick). When the narrator adopts a tone that allows them to distance themselves from information given conditionally, for example, ChatGPT translates in the conditional instead of using expressions like 'is said to be,' 'allegedly,' or 'reportedly,' etc.
-
ChatGPT is not eloquent in translation. It produces too much literal and word-for-word translation. It does not engage in free translation. It sticks too closely to the structure of the source text.
-
It does not perform oblique translation, such as modulation or change of perspective, transposition, equivalence, and transediting.
V. DISCUSSION
5.1 Syntactic Errors
1st Example
| ST: Tous les chefs portaient le titre de Okoriko jusqu'au grand-père de celui qui est encore au pouvoir. (p.43) | TT: All chiefs bore the title of Okoriko until the grandfather of the current one in power. | Correction: All chiefs, including the grand father of the incumbent chief, bore the title of Okoriko. |
ChatGPT's translation is literal and meaningless. In this sentence, the subject of the verb is "All chiefs, including the grandfather of the incumbent chief," but ChatGPT's translation does not show that the phrase "jusqu'au grand-père de celui qui est encore au pouvoir" is part of the subject. It should be noted that this phrase has been restructured; indeed, "jusqu'au grand-père de celui qui est encore au pouvoir" is incorporated into the sentence before the verbal group "portaient le titre d'Okoriko." Thus, this phrase becomes an integral part of the subject of the verb "portaient." In summary, the subject of the verb "portaient" is "Tous les chefs... jusqu'au grand-père de celui qui est encore au pouvoir," and the predicate is "portaient le titre d'Okoriko." ChatGPT's translation contains a serious structural error that shows a lack of understanding of the meaning of the sentence as well as the grammatical functions of the different phrases within the sentence.
2nd Example
| ST: Le chef de cette maison procède au cérémonial il rampe, tenant une calebasse contenant de l'eau qu'il remet au nouveau roi pour se débarbouiller. (p.38) | TT: The chief of this house performs the ceremonial act of crawling, holding a calabash containing water which he hands over to the new king to cleanse himself. | Correction: In performing the ceremony, the chief of this household crawls, while holding a calabash containing water which he hands over to the new king to wash his face. |
ChatGPT does not perform transediting; otherwise, it would have improved the structure of the translation, which is incorrect because the
French text is also incorrect. Note that the revised translation replaced the pronoun 'himself' with 'his'.
Third Example
| ST: Le conseil donna raison à cet homme venu de Savè et élevé dans la tradition baatonu qui voulait que le plus grand occupât la première place. (p.35) | TT: The council sided with this man from Savè, raised in the Baatonu tradition, who wanted the eldest to occupy the first place. | Correction: The cabinet sided with this man from Savè, raised in the Baatonu tradition, which wanted the eldest to occupy the first place. |
The relative pronoun 'who' used in the Target Text (TT) is incorrect because it refers to "the man from Savè." The pronoun to be used should represent the term "Baatonu tradition."
Therefore, 'who' is replaced by 'which.' This indicates that ChatGPT is unaware of the cultural context of the Baatonu.
4th Example
| ST: A chaque sortie officielle, il était obligé de suivre à cheval ses jeunes frères dont les griots chantaient obligatoirement les louanges. (p.34) | TT: On every official outing, he was obliged to follow his younger brothers on horseback, whose praise was sung by griots. | Correction: On every official outing, he was obliged to follow on horseback his younger brothers whose griots compulsorily sang their praises. |
There is a structural change in the phrase 'whose praise was sung by griots' involving the introduction of the adverb 'compulsorily.' This adverb was omitted by ChatGPT. The plural form of 'praise' (i.e., 'praises') is used. Note that
ChatGPT translates: "ses jeunes frères dont les louanges étaient chantées par les griots," but the revised translation is: "ses jeunes frères dont les griots chantaient obligatoirement les louanges."
Fifth example
| ST: <<O mon na O bourou bèè» (p.33) | TT: "O my na O bourou bèè" | Correction: <<O mon na O bourou bèè» |
ChatGPT failed to recognize a vernacular expression and translated only the word that resembles a French word. The result "O my na O bourou bèè" is meaningless. The word "mon" in the nago language was translated into English, which is incorrect. Normally, this vernacular expression should not be translated.
Sixth Example
| ST: Ces derniers étaient soumis à des taxes de douane au village de Wolou. (p.32) | TT: These were subject to customs duties in the village of Wolou. village on which Tourou depended. | Correction: They had to pay customs duties in the village of Wolou. |
Here is another translation: 'They had to pay customs duties.' Referring to the Hausa traders, ChatGPT used 'these' instead of 'they' or 'the latter.' ChatGPT's translations are often 'out of context' and literal.
7th Example
| ST: C'est un grand détenteur de gris-gris qui devient Baba Damagui. (p.30) | TT: He is a great holder of talismans who becomes Baba Damagui. | Correction: Only a great holder of supernatural powers becomes Baba Damagui. // An alternative translation: 'To become Baba Damagui you have to be a powerful juju man.' |
"ChatGPT's translation is incorrect. It is not at all faithful to the source text. That is why a change in sentence structure was deemed necessary to replace the phrase 'he is' with the word 'only'.
8th Example"
| ST: Le Kpébié Sounon (ou chef de Kpébié) est issu de Nikki par sa mère, du clan Yari Sannou à Danrì. (p.29) | TT: The Kpébié Sounon (or chief of Kpébié) is descended from Nikki through his mother, from the Yari Sannou clan in Danri. | Correction: The Kpébié Sounon (or chief of Kpébié) hails from Nikki; indeed, his mother is from the Yari Sannou clan in Danri. |
"ChatGPT's translation is literal. The translation would be more understandable if the following structure were used: 'The Kpébié Sounon (or chief of Kpébié) hails from Nikki; indeed, his mother is from the Yari Sannou clan in Danri.' The revised sentence is clearer. The adverb 'indeed' confirms the assertion made in the main clause. This is an example of structural change. One clause is transformed into two.
5.2 Semantic Errors
1st Example"
| ST: Origine de la société politique wassangari du Baru Wu (pays Baatonu). (p.16) | TT: Origin of the Wassangari political society of Baru Wu (Baatonu country). | Correction: Origin of the Wassangari political society of Baru Wu (Baatonu homeland). |
"The synonym 'homeland' is preferred because the word 'country' generally refers to a sovereign territory with defined borders, a central government, and institutions. This is not what the author means by 'pays Baatonu.' He is referring to a territory that encompasses regions and areas within a larger country, and a place that often and identity. This type of semantic error is carries a personal sense of attachment, belonging, common in ChatGPT's translation.
2nd Example" "Misunderstanding of the meaning of the word 'mot' in this context where it means 'remarks'."
| ST: C'est par ce mot de Jacques Lombard que je voudrais introduire ces débats sur les éléments d'anthropologie et d'histoire du Nord-Bénin. (p.5) | TT: It is through this word of Jacques Lombard that I would like to introduce these debates on the elements of anthropology and history of Northern Benin. | Correction: It is through these remarks by Jacques Lombard that I would like to introduce this discussion on anthropological and historical elements of Northern Benin.The word 'mot' is literally translated by ChatGPT, but in this context, it means remarks. This translation error is both semantic and pragmatic. This sentence comes after a series of three remarks by Jacques Lombard; that's why the word 'remarks' is used in replacement of the word 'word'. |
3rd Example" "The literal translation 'express theft' by ChatGPT is replaced by 'stealing on purpose'. In fact, 'express theft' means 'vol express' in French.
| ST: «Vol exprès»: à la naissance d'un enfant de ce clan, on doit aller «voler>> tout le nécessaire pour la toilette du bébé; mais on doit d'abord aviser les gens chez qui on va <<voler ces éléments, d'où les louanges de «Kannou Gbinnou» ou «Kinnou-voleur». (p.9) | TT: "Express theft": upon the birth of a child from this clan, one must go "steal" all the necessary items for the baby's care; but one must first inform the people from whom these items will be "stolen", hence the praises of "Kannou Gbinnou" or "Kinnou-thief". | Correction: "Stealing on purpose": upon the birth of a child from this clan, someone must go and "steal" all the necessary items for the baby's care; but the people from whom these items are "stolen" are informed in advance, hence the praises of "Kannou Gbinnou" or "Kinnou thief". |
4th Example"
| ST: Les Wonko sont réputés pour leur puissance sur le plan occulte. (p.9) | TT: The Wonko are renowned for their power in the occult realm. | Correction: The Wonko are renowned for their supernatural powers. |
"The term 'supernatural' is a hyponym of the hypernym). By using it, the reader will better expression 'in the occult realm' (which is the understand what it is about.
5th Example"
| ST: Les Baatombu font la distinction entre les assimilés lointains et les assimilés récents qui se réclament tous de la même appartenance bariba. | TT: The Baatombu make a distinction between distant assimilates and recent assimilates who all claim the same Bariba membership. | Correction: The Baatombu make a distinction between distant assimilates and recent assimilates who all claim to be members of the same Bariba ethnic group. |
6th Example"
| ST: dans le sens moins large, il pourra signifier l'ensemble culturel formé par différents groupes ethniques, déjà fortement intégrés, participant au système politique et ayant des genres de vie proches; et il faudra alors exclure les groupements assujettis ou étrangers: Gando, Peul et Dendi des caravansérails; | TT: In a narrower sense, it may signify the cultural whole formed by different ethnic groups, already strongly integrated, participating in the political system and having similar ways of life; and then it will be necessary to exclude the subject or foreign groups: Gando, Peul, and Dendi from the caravanserais; | Correction: In a narrower sense, it may signify the cultural whole formed by different ethnic groups, already strongly integrated, participating in the political system and having similar ways of life; and then it will be necessary to exclude the dominated or foreign groups: Gando, Peul, and Dendi from the caravanserais; |
"By using the adjective 'dominated,' the corrected translation expresses a more concrete idea. 'Subject groups' does not accurately translate the term 'groupements assujettis.' This change, referred to as 'abstract change' by Chesterman, provides more precision and makes the correction more accurate.
5.3 Pragmatic Errors
1st Example
| ST: L'expression suivante est assez significative pour caractériser les Booro: «Sou daara wa, sou kou werou wa, Booro», ce qui, traduit, veut dire «qu'on voit l'aller et qu'on ne voie pas le retour, Booro». En d'autres termes, lorsqu'un Booro voyage, la manière de lui souhaiter bon voyage, agréable séjour et bon retour, c'est de lui dire «qu'il perisse là-bas et qu'il ne revienne jamais». | TT: The following expression is quite significant in characterizing the Booro: "Sou daara wa, sou kou werou wa, Booro", which, translated, means "let them see the departure and not the return, Booro". In other words, when a Booro travels, the way to wish them a good journey, a pleasant stay, and a safe return is to say "may they perish there and never return". | Correction: The following expression is quite significant in characterizing the Booro: "Sou daara wa, sou kou werou wa, Booro", which means "go and never come back Booro". In other words, when a Booro travels, the best way to wish them a safe journey, a pleasant stay, and a safe return is to say: "May you perish there and never come back". |
"The modification made in this section is an example of an 'illocutionary change.' Indeed, the mood of the expression in italics in the revised translation is imperative, which marks a change from the source text, which is in the simple present tense. This change in mood is the best way to faithfully convey what the author expresses through the explanation following the expression.
2nd Example"
| ST: Certaines populations de Chabı Kouma (dans l'Atacora), se sont fusionnées dans ce clan, ce qui a fait soutenir par certains auteurs son origine «atacorienne».(p.11) | TT: Some populations of Chabı Kouma (in Atacora) merged into this clan, which led some authors to support its "atacorian" origin. | Correction: Some people from Chabı Kouma (in the Atacora region) merged into this clan, which led some authors to confirm that its origin is indeed in the Atacora region. |
"The adjective 'atacorian' is rendered by the noun 'Atacora' in the correction. This is an example of transposition. The adjective 'atacorian' is not a formal word that can be found in a dictionary. It is a word coined for the occasion.
3rd Example
| ST: Ainsi, contrairement aux membres des clans autochtones qui sont alignés sur le côté paternel pour leur appellation, les Wassangari sont différenciés par leur côté maternel. Mais pour la succession au trône, c'est le patriarcat qui est adopté. Ainsi, tous les clans Wassangari sont ceux de leur côté maternel. | TT: Thus, unlike members of indigenous clans who are aligned on the paternal side for their appellation, the Wassangari are differentiated by their maternal side. However, for succession to the throne, patriarchy is adopted. Thus, all Wassangari clans are those of their maternal side. | Correction: Thus, unlike members of indigenous clans who are named after their father, the Wassangari are different because they are named after their mother. However, as far as succession to the throne is concerned, patriarchy is adopted. Thus, in all Wassangari clans, people are named after their mother. |
"The current paragraph has undergone several modifications to make it clear for the reader of the target text (TT). In addition to the syntactic change, the final sentence has been revised to ensure that the reader understands its exact meaning.
4th Example"
| ST: Ils ont su, par certaines qualités exceptionnelles, usurper le pouvoir politique des autochtones pour devenir des éléments incontournables des sociétés qui les ont intégrés. (p.12) | TT: They have managed, through certain exceptional qualities, to usurp the political power of the indigenous people and become indispensable elements of the societies that have integrated them. | Correction: They have managed, owing to their exceptional qualities, to usurp the political power of the indigenous people and become indispensable elements of the societies that have integrated them. |
"A pragmatic change: 'through certain exceptional qualities' is replaced by 'owing to their exceptional qualities.' Note that 'certain exceptional qualities' is replaced by 'their exceptional qualities.' The pronoun 'their' is a deictic reference specifically to the Wassangari. ChatGPT produced a literal translation that is meaningless. However, it is also important to note that the French text is not clear either.
5th Example
| ST: A l'origine, leur chef serait le premier fondateur du Mali. Ils sont souvent musulmans. (p.13) | TT: Originally, their chief would be the first founder of Mali. They are often Muslims. | Correction: Historically, their chief was said to be the first founder of Mali. They are often Muslims. |
"The present conditional used in the target text (TT) is replaced by the phrase 'was said to be' in the correction. The author is not sure of the information they provide. The modal 'would' is not the most appropriate element to express the distance the author takes vis-à-vis their statement in this context. The phrase 'was said to be' better expresses the cautious position the author takes regarding the information they provide.
6th Example
| ST: Pour se faire, l'origine, la structure et le fonctionnement institutionnel de la société politique wassangari constituent les éléments fondamentaux sur lesquels doivent s'ouvrir des débats qui apporteront, nous l'espérons, un éclairage à la vie du peuple baatonu. (p.16) | TT: To do this, the origin, structure, and institutional functioning of the Wassangari political society constitute the fundamental elements on which debates must be opened, which we hope will shed light on the life of the Baatonu people. | Correction: To this end, the origin, structure, and institutional functioning of the Wassangari political society constitute the fundamental elements of a discussion that will hopefully shed light on the life of the Baatonu people. |
"The expression 'pour se faire' is not well written. It should be 'pour ce faire.' This is an example of transediting, according to Chesterman. Moreover, the interpretation of the expression 'pour ce faire' in this context is incorrect. In reality, this expression means 'to this end' in this context. Translating 'pour ce faire' as 'to do this' does not match the meaning it has in the target language. Indeed, English speakers often say 'to this end,' which means 'by working towards this goal.'
7th Example
| ST: Ces littératures laissent cette fausse impression qu'il n'y avait rien d'organisé dans le Baru Wu avant les Wassangari. (p.16) | TT: These literatures leave the false impression that there was no organization in Baru Wu before the Wassangari. | Correction: This literature gives the false impression that there was no form of organisation in the Baru Wu before the Wassangari's arrival. |
"Note the change from 'before the Wassangari' to 'before the Wassangari's arrival.' Indeed, in this particular context, the arrival of the Wassangari is what the author is referring to.
VI. CONCLUSION
ChatGPT contains a vast number of words, but it is not the quantity of words that a translator knows that makes them a good translator. This syntactic, semantic, and pragmatic evaluation of ChatGPT's translation has yielded disappointing results, leading us to say that, at its current stage, this translation tool is merely a translation aid, even though it is more advanced than other tools such as Systran, Google Translate, Microsoft Translator, etc. The concerns of translators about the future of their profession are not justified unless humanity insists on using machine translations, which will cause twice as much harm as good.
It should be noted that throughout human history, translation errors have caused tragedies such as the bombing of Hiroshima and Nagasaki, deaths in operating rooms, bank failures, aeroplane crashes, and many others.
8. https://chat.openai.com/.
Conflict of Interest
The authors declare no conflict of interest.
Ethical Approval
Not applicable
Data Availability
The datasets used in this study are openly available at [repository link] and the source code is available on GitHub at [GitHub link].
Funding
This work did not receive any external funding.
References
Cite this article
Related Research
Special Issue
Launch a focused special issue to highlight research, emerging trends, and expert insights in your academic field.
