Introduction
In recent years, artificial intelligence in education (AIEd) has gained increasing relevance. AIEd promises adaptive and personalised learning environments, with potential ranging from reducing the mundane workload for both students and teachers to creating more engaging educational experiences (Rivoltella and Rossi Reference Rivoltella and Rossi2019).
However, this innovation requires an approach to promote the use of artificial intelligence (AI) to develop students’ critical thinking skills, i.e., foster critical thinking and intellectual autonomy, and to ensure ethical use aligned with social well-being and human-centred values (Han Reference Han2018).
Recently, this area has been revolutionised by the emergence of generative artificial intelligence (GAI). Tools like ChatGPT (OpenAI 2024) are also increasingly used in secondary education. The integration of GAI in secondary education, particularly for the teaching of ancient languages, remains a relatively underexplored field, partly due to past and, in some cases, ongoing resistance to its adoption. Recent literature has seen the emergence of some studies and proposals beginning to explore this area. For example, Ross (Reference Ross2023) discussed various theoretical applications of ChatGPT, with a particular focus on its use as a virtual tutor. Díaz-Sánchez and Chapinal-Heras (Reference Díaz-Sánchez and Chapinal-Heras2024) proposed the use of Midjourney, an image generator, to support students during translation tasks. Moreover, it has been demonstrated that the integration of a Latinism classifier, an ELECTRA model specifically trained for this purpose, with a Latin text generator, a similarly fine-tuned GPT-2, can assist students in engaging with various linguistic concepts (Cavaleri Reference Cavaleri2021, Reference Cavaleri2022). Peddar adopts a proactive approach to the use of technologies and GAI tools in a Latin classroom and develops strategies to encourage students’ reflection, in order to enhance their understanding of the uses and misuses of GAI platforms in their learning (Peddar Reference Peddar2025).
The introduction of GAI in secondary education, particularly language models such as ChatGPT and DeepSeek-R, represents both a challenge and an opportunity to rethink traditional teaching methodologies and can make the study of Latin more dynamic, accessible, and engaging. Unlike traditional search engines, GAI offers original solutions, simulated dialogues, personalised grammatical explanations, and translations with stylistic commentary. However, some studies, such as (Lee et al. Reference Lee, Sarkar, Tankelevitch, Drosos, Rintel, Banks and Wilson2025), argue that AI reduces the effort required for critical thinking and that automated feedback often overlooks affective and motivational dimensions, which are crucial for learning. On the other hand, as Law (Reference Law2025) points out, critical thinking is not solely based on effort but on strategy. Indeed, if GAI allows us to automate routine tasks, this does not mean we think less; it simply means that our way of thinking is changing, and this shift can represent an opportunity rather than a loss, provided that we learn to use GAI appropriately.
In this paper, we explore the adoption of GAI in an Italian Liceo Classico (Classical High School), ‘Liceo Lucrezio Caro’ of Rome. Traditionally, the Liceo Classico has been considered the type of high school that best prepares students for university studies, particularly in the humanities, law, history, and philosophy. Its main characteristic is a broad and rigorous cultural education, with a strong emphasis on the study of classical languages and civilisations (Latin and Greek), as well as philosophy, literature, history, and scientific subjects.
Currently, there is resistance to the adoption of GAI tools in Liceo Classico. However, such resistance may prove futile and potentially detrimental in the long run. Rather than rejecting GAI a priori, educators should explore the possibility of embracing it. Mindfully integrated, GAI has the potential to enhance, not diminish, the study of Latin and Greek as part of the educational journey. A good starting point for the effective use of this new tool is grounding it in constructivist learning theory. Rooted in the works of Dewey, Piaget, and Vygotsky, constructivist learning theory promotes the active and autonomous construction of knowledge (Bruner, Reference Bruner1961; Dewey Reference Dewey1929; Piaget Reference Piaget and Piatelli-Palmarini1980; Vygotsky Reference Vygotsky1962). GAI can support this process through personalised dialogue and individual feedback, supporting learning without replacing the teacher’s role. The current resistance to GAI can be compared to the initial reluctance towards calculators in mathematics education. Indeed, calculators changed teaching approaches, but not in a negative way. Similarly, GAI could become a kind of ‘calculator’ for the humanities, a study aid that does not replace the teacher but complements them (Bozkurt et al. Reference Bozkurt, Xiao, Lambert, Pazurek, Crompton, Koseoglu, Farrow, Bond, Nerantzi, Honeychurch, Bali, Dron, Mi, Stewart, Costello, Mason, Stracke, Romero-Hall, Koutropoulos, Toquero, Singh, Tlili, Lee, Nichols, Ossiannilsson, Brown, Irvine, Raffaghelli, Santos-Hermosa, Farrell, Adam, Thong, Sani-Bozkurt, Sharma, Hrastinski and Jandrić2023), it augments their expertise.
It is clear that GAI has limitations: linguistic irregularities, for example, when GAI blends ancient and modern languages, as in ancient Greek (Ross Reference Ross2023); grammatical inaccuracies, as GAI models often struggle to reliably explain grammar concepts due to insufficient training data (Bendel and N’diaye Reference Bendel and N’díaye2023); difficulty recognizing linguistic variation, as GAI may fail to distinguish different meanings or translations in morphologically similar forms, as in Latin and Greek languages (Ross and Baines Reference Ross and Baines2024); language confusion and hallucinations, particularly when mixing Latin-derived or Latin-script languages (Bistafa Reference Bistafa2023); and content restriction issues, as GAI may refuse to display texts on culturally relevant topics such as gender roles, sexuality, war, and slavery, thereby impacting classical studies (Cavaleri et al. Reference Cavaleri, Hersch and Kolde2025; Ross Reference Ross2023). This is why it is essential for students to be equipped with the necessary skills to navigate this sea of information with awareness, to understand whether it is truthful or manipulated and distorted by more or less malevolent algorithms (Ranieri et al. Reference Ranieri, Cuomo and Biagini2023). Indeed, it is fundamental to consider these weaknesses of GAI, but it is also fundamental to consider its strengths.
So, this study investigates the application of GAI in the teaching and learning of Latin within upper secondary education, with a particular focus on the Italian Liceo Classico, a context still underexplored in current literature. By analysing classroom activities conducted during the project, the research contributes to broader discussions on how GAI is reshaping educational practices, providing empirical evidence of its potential to enhance Latin language skills and student engagement. Through a dual focus on both learning outcomes and student experiences, this study offers a comprehensive assessment of the role of GAI in linguistic development and classroom participation. It highlights not only the opportunities offered by this technology but also its limitations, helping to identify best practices for its integration into secondary school contexts.
The project
The project, called ‘AI and Latin: Technological Innovation in the Service of Classical Tradition’, was conducted through a collaboration of Sapienza University of Rome and the Liceo Lucrezio Caro in Rome, within the Italian Ministry of Education’s PCTOFootnote 1 scheme. It consisted of five in-person modules (two hours each) and an additional five hours of independent work assigned as homework. These modules were taught in addition to the daily hours. The activities were carried out between February and March 2025 and involved two third-year classes.
The project aims to develop in students the ability to use GAI tools to improve the learning of Latin language and culture; foster critical dialogue on translation and linguistic choices; improve the ability to analyse texts in a more in-depth way; develop the ability to formulate sentences in Latin, while improving the understanding of everyday vocabulary and basic grammar; and stimulate imagination and encourage the active use of Latin, bringing out grammatical and syntactic aspects in a natural way to enhance creativity and critical thinking
Preliminary activities
Since there was no direct knowledge of the students, some preliminary activities were carried out: a meeting with the Latin subject teachers and the tutor teachers to gain a general understanding of the class’s relationship with Latin, the use of GAI tools, and their linguistic knowledge and skills; a background questionnaire; and a questionnaire on GAI.
The meeting with the class teachers and tutors allowed me to understand the class level and the topics covered in the third year during the first term. The outcome of the meeting was that the students in both classes have roughly the same level of knowledge of the Latin language, which is the one expected from a third-year class of a good Liceo Classico, and the results of both classes are similar. The same applied to Latin literature, whose knowledge in both classes was at a medium level, and certain aspects, such as the style of Plautus and Cicero, have not been studied deeply yet.
The results of the knowledge survey are: 1) General profile of students: Most students report being born in 2008, making them around 16–17 years old. The gender is predominantly female, and all state that they speak Italian at home as their primary language, indicating a rather homogeneous linguistic background; 2) Use of GAI: Most students are familiar with and use GAI systems, particularly ChatGPT. The main uses suggest a growing integration of GAI into daily school practices. GAI is used for a) asking questions; b) deepening topics; c) summarizing texts; and d) supporting studying. 3) Awareness and critical attitudes: Many students report engaging in conversations with GAI systems and do not limit themselves to receiving a single answer. However, an interesting finding is that only a part of the students verifies the correctness of the answers obtained, for example, by cross-checking information on other sites or with reliable sources. This indicates that there is room to enhance critical digital competence; 4) Perception of usefulness and risks: a) Usefulness: The majority of students believe that GAI is a valuable aid for studying, as it allows them to clarify doubts, conduct research, or summarise texts; and b) Risks: A minority, but a significant portion, fears an intrusive or passivating use of GAI, potentially hindering independent learning.
The general conclusions on the results of the incoming knowledge survey are: the analysis of student responses reveals an overall positive and conscious attitude towards the use of GAI; almost all students are already using GAI tools for studying. The responses show a good tendency towards active interaction with GAI systems; however, since only a minority report checking the information obtained, further work is needed to develop critical thinking and more thoughtful approach to digital skills; GAI is mostly seen as a useful and valuable tool for studying, though there are more cautious views that point to the risk of excessive or passivating use, potentially detrimental to their autonomy as students; in summary, the two classes involved in the survey showed interest and openness towards the use of GAI, recognizing its educational potential.
First module: introduction to GAI
The module work was designed to be collaborative and exploratory, encouraging active student engagement.
The module started with the instructor giving a general overview of the key concepts of GAI. Students were then asked to share their prior experiences using GAI tools.
Following this, they worked in pairs to search and define key terms related to the World Wide Web and Artificial Intelligence. This search activity was conducted using both traditional methods (e.g., consulting informative websites), and with GAI tools.
Then, a class-wide discussion was held focusing on two main topics: the concept of ‘prompting’ and the responsible use of GAI.
After the discussion, the instructor gave indications about what is prompting, how to do it, and the responsible use of GAI. We report such indications below.
What is prompt engineering and how to prompt
Definition: Prompt Engineering is the process of crafting clear and specific instructions that guide the GAI language model in generating relevant and accurate responses.
The goal is to ensure that the generated responses meet your requests. Therefore, it is important to refine the skill of asking the right questions and creating precise prompts. This ensures that the GAI language model provides answers that are relevant to specific needs and learning objectives.
Guidelines for effective prompting: To create effective prompts, students were given the following set of guidelines to use with a GAI tool (ChatGPT-4o):
-
• Identify the specific purpose of the prompt: Determine the main reason for writing the prompt.
Example: a prompt to generate creative writing or to translate a text from Latin into Italian.
-
• Understand the audience’s objective: Know the audience’s age, learning level, and/or cultural background in order to create prompts that are appropriate and engaging.
-
• Use clear and concise language: Use precise and understandable language so the GAI can easily follow the instructions. This helps the GAI language model generate accurate and precise responses.
-
• Provide contextual information: Be sure to include the necessary information to help the GAI language model generate accurate and relevant answers. This includes introducing the topic, providing examples, or linking to additional resources.
Students were also given ‘The Principles of Prompt Engineering’ (AI for Education and Cubero Reference Cubero2023):
-
• Provide context about the role, expertise, and/or environment for the expected output
Giving context about the role, expertise, and/or environment means offering specific instructions to obtain the correct output.
Example: You are a student in classical languages, specifically a third-year Italian High school (Liceo Classico) student.
-
• Clearly specify the task and give details on what to include
Prompt instructions should clearly define the task and express in detail what should be included.
Example: You are a third-year high school student (Liceo Classico). Prepare a 60-minute lesson on the five Latin declensions. You must provide examples and diagrams that highlight the features of each declension, including irregularities.
-
• Simplify the language and provide examples
Using a conversational approach with simple and concise language and providing examples helps the GAI better understand what you’re looking for. A prompt can include relevant information on the topic, examples, and links to additional resources. Even a single example can significantly improve the quality of responses.
Example: Explain the first Latin declension. Include a practical example, such as a Latin sentence with analysis (case, gender, number, if it’s an irregularity) of the words from the first declension and a translation of the sentence.
-
• Structure the output – tell the chatbot how to format the response, the target audience, and/or the sections
Tell the chatbot how to organise the response, with specifics on the format, the intended audience, and/or the sections.
Example: Create a table of Latin verb rules for me (I am a high school student of Liceo Classico and I need an easy way to repeat the rules), formatted as a table divided into sections.
-
• Provide feedback
Give feedback throughout the conversation. Specify what needs to be changed to better meet your needs.
Responsible use of GAI
The instructor gave some ‘instructions’ on responsible use of GAI, taking advantage of the acronym ‘EVERY’ proposed in (AI for Education and Cubero Reference Cubero2023). Each letter of this word composes some key-sentences, so in this way it is easier to remember them: EVALUATE the initial output to see if it meets the intended purpose and your needs; VERIFY facts, figures, quotes, and data using reliable sources to ensure there are no hallucinations or bias; ENGAGE in every conversation with the GenAI chatbot, providing critical feedback and oversight to improve the GAI’s output; REVISE the results to reflect your unique needs, style, and/or tone. GAI output is a great starting point, but shouldn’t be a final product; YOU are responsible for everything you create with GAI. Always be transparent about how you’ve used these tools.
The latter was demonstrated through practical examples (a Latin text); students used GAI tools to generate explanations of Latin morphosyntax, perform syntactic analysis, attempt translations, and explore other applications relevant to the study of Latin texts. This showed to students the potential of GAI as a powerful learning aid in classical language education.
To reinforce the content of the module and promote autonomous learning, students were assigned a search task to be completed at home. They were asked to investigate three foundational topics: neural networks, the Turing Test, and potential errors and limitations of GAI.
Home assignment: As homework, they were instructed to apply basic prompting strategies in the analysis of a Latin text, thereby integrating classical studies with modern technological tools. Specifically, the following assignment was given:
-
• Analyse and translate the following text from Latin into Italian.
-
• After completing the analysis and translation of the text below, reread the handouts on prompt engineering.
-
• Compare your own translation with the one provided by the GAI (ChatGPT, Copilot, Gemini, etc.); it is also possible to compare translations from different GAI tools. Check which parts of the translation are incorrect (attention! Remember that GAI can also make mistakes) and write your own reflections on the translation and the errors you found.
-
• Identify the grammatical rules that need to be reviewed by asking the GAI to clarify them (again, check with your grammar book to ensure the answers are correct).
-
• Formulate prompts to ask the GAI for further information about characters present in the text or other related aspects.
-
• Transcribe on Google Classroom your own translation of the text below and the GAI’s translation, highlighting the differences and explaining them, as well as all the prompts and the corresponding answers, pointing out any errors made by the GAI (if any).
The future anticipated by dreams (Cicero)
Matrem Phalaridis scribit Ponticus Heraclides, doctus vir, auditor et discipulus Platonis, visam esse videre in somnis simulacra deorum, quae ipsa domi consecravisset; ex iis Mercurium e patera, quam dextera manu teneret, sanguinem visum esse fundere; qui cum terram attigisset, refervecere videretur sic, ut tota domus sanguine redundaret. Quod matris somnium immanis filii crudelitas comprobavit. Quid ego, quae magi Cyro illi principi interpretati sint, ex Dinonis Persicis libris proferam? Nam cum dormienti ei sol ad pedes visus esset, Dinon scribit ter eum frustra adpetivisse manibus, cum se convolvens sol elaberetur et abiret; ei magos, quod genus sapientium et doctorum habebatur in Persis, dixisse ex triplici adpetitione solis triginta annos Cyrum regnaturum esse portendi. Quod ita contigit.
The initial findings confirmed the results of the preliminary questionnaire: most students were already familiar with GAI, particularly ChatGPT. However, they had little to no prior knowledge of prompt engineering, and their use of GAI was not consistently dialogic or reflective of effective conversational interaction with the tool, an inconsistency with what they had self-reported in the questionnaire. Moreover, they did not know how the GAI works, and they were not familiar with ‘hallucinations’.
Second module: linguistic and semantic applications
This module involved individual translation activities aimed at acquiring knowledge related to students’ preparation on certain grammatical and syntactical rules, as well as lexical knowledge. The aim is to have a basis to later assess whether the lessons within the project had a positive impact on the students’ Latin language acquisition, by assigning and marking a similar translation at the end of the project, featuring comparable rules and vocabulary.
Indeed, based on data collected by the Latin teacher concerning the students’ knowledge and skills, the instructor assigned a medium-level Latin text for translation, to be carried out using a Latin dictionary. The selected passage, by Cassius Dio’s adapted work ‘Ρωμαϊκὴ ἱστορία, discusses the emperor Commodus and includes mainly terms related to the vocabulary of potestas and educatio, as well as syntactic rules involving temporal, causal, concessive, and relative clauses.
The results showed that of 41 students, only 19 received a good mark. This result is in line with what is expected from this kind of translation.
Home assignment: a home assignment was given based on the version on Commodus completed in class and students had to apply what they had learned. The assignment was the following:
-
• Identify in the given text the terms that belong to the semantic fields of education and potestas. Ask the GAI, using the correct prompt, to do the same exercise. Compare the GAI’s response with your work and integrate it, if necessary.
-
• Identify the terms that are still present in our language. Ask the GAI to complete this exercise and compare it with your work. Remember that GAI results may contain hallucinations.
Commodus (text adapted by Cassius Dios’s ‘Ρωμαϊκὴ ἱστορία)
Commodus, ut primum potestate potitus est, neque legibus neque magistratibus obtemperavit, sed sua sponte solum imperatorem esse voluit, omnemque educationem suam ad ludos et divitias contraxit. Nam cum in pueritia sua laudabilis esset et magistris, quos pater eius ad eum destinaverat, obtemperaret, postea contra leges optimos viros secutus est. Verum quia idem non ad adamodum virtutem sectabatur, sed improbitatem et luxuriam pervagavit, non solum se ipse, sed multos etiam alios in efferatos mores erudivit. Nam cum adulescentiam egens virtutis omnimodae, quae bonis et pulchris moribus superaret, sensisset, si quid verum etiam aut pulchrum esset, id omne aut pro nihilo habebat aut gravissime probro inter se ipsos ferebat.
Third module: comparison of human translation and GAI translation
This module was based on a comparison between human and GAI-generated translation.
The motivations were:
-
• to highlight similarities, differences, and limits;
-
• to reflect together with students on how prompting works and how linguistic, interpretive, and stylistic choices can vary between humans and machines;
-
• to develop critical awareness of using GAI in classical language study: understanding when and how GAI can support learning;
-
• to address the issue of ‘hallucinations’ (errors or misleading content generated by GAI);
-
• to promote independent verification practices.
This was a collaborative activity.
The Cassius Dio passage on Commodus had already been completed previously in the second module, and its analysis was assigned as homework.
In class, the GAI was tasked with analysing and translating the text itself, allowing observation of how to build an automatic translation from a known text.
Then, there was a collective discussion: after the GAI translation, students discussed in groups the prompting strategies used, comparing approaches and results.
Later, there was a guided correction together with the students, comparing their version with the GAI one, to highlight differences, errors, and good practices.
A final discussion highlighted the difficulties encountered and how human and GAI translations can diverge.
Through this module, students understood that GAI can offer a useful base of work, but it is not infallible. They started to develop critical evaluation skills: when automatic translation is reliable and when it requires verification. Indeed, they learnt effective prompting practices, i.e., the instructions to give the GAI to obtain more useful translations and to explore different interpretations.
In brief, this guided lab activity consisted of a collaboration between students and GAI to translate a Latin passage, followed by a critical discussion about translation processes, differences between humans and machines, and the importance of recognizing and managing ‘hallucinations’, to use GAI as an educational support, not a substitute for the teacher.
We give below some details of the activities performed.
For example, we asked GAI to translate the following sentence of the previous Latin text titled ‘Commodus’:
Prompt: Translate the following Latin text: Commodus, ut primum potestate potitus est, neque legibus neque magistratibus obtemperavit, sed sua sponte solum imperatorem esse voluit, omnemque educationem suam ad ludos et divitias contraxit.
ChatGPT Translation: Commodus, as soon as he gained power, did not obey either the laws or the magistrates, but wanted to be the sole emperor and reduced all his education to games and wealth.
Student translation: Commodus, when he took power, obeyed neither the laws nor the magistrates, but of his own will wanted to be the only emperor, and directed all his education toward games and riches
We can observe, comparing the two translations by the students and by GAI, that GAI did not translate ‘sua sponte’ (= of his own will). So, GAI is not precise.
As another example of analysis and translation, we asked to ChatGPT:
Prompt: Analyse the following Latin sentence grammatically and syntactically by identifying the clauses, and then translate it litterally: Nam cum in pueritia sua laudabilis esset et magistris, quos pater eius ad eum destinaverat, obtemperaret, postea contra leges optimos viros secutus est.
ChatGPT answer:
Analysis:
Causal conjunction: Nam (introduces an explanation or justification).
Temporal clause: cum in pueritia laudabilis esset (imperfect subjunctive: ‘when he was praiseworthy’).
Concessive clause: et magistris obtemperaret (imperfect subjunctive: ‘and obeyed the teachers’).
Main clause: postea contra leges optimos viros secutus est (perfect indicative: ‘afterwards he followed noble men against the laws’).
Translation: For although in his childhood he was praiseworthy and obeyed the teachers, whom his father had carefully chosen, later he followed noble men against the laws.
We can notice that:
-
• There isn’t a complete analysis of the sentence.
-
• Nam: isn’t a causal conjunction, but it is an explanatory conjunction
-
• Cum: in this sentence, it doesn’t introduce a temporal clause, but a concessive one.
-
• ChatGPT doesn’t analyse the clause: ‘quos…destinaverat’.
-
• ChatGPT added in the translation of ‘quos…destinaverat’ the adverb ‘carefully’, but it isn’t in the text.
Indeed, we asked to ChatGPT:
Prompt: Is ‘cum in pueritia sua laudabilis esset et…obtemperaret’ a concessive clause?
Note that students know that this is a concessive clause, and ChatGPT translated it as a concessive clause, but ChatGPT responded incorrectly in its analysis that the sentence is a ‘temporal clause’.
ChatGPT answered:
Excellent question.
The clause: cum in pueritia sua laudabilis esset et magistris… obtemperaret …is introduced by “cum” and contains subjunctive verbs:
laudabilis esset (imperfect subjunctive of ‘esse’)
obtemperaret (imperfect subjunctive of ‘obtemperare’)
This construction can be temporal, causal, or concessive, depending on the context.
Is it concessive?
Yes – in this context, the clause is best interpreted as concessive.
Why?
The main clause tells us that he later followed the best men against the laws – a negative or rebellious act.
The cum-clause tells us that in his childhood, he was praiseworthy and obedient – a positive description.
So, there’s a contrast between:
He was praiseworthy and obedient in childhood versus
Later he followed men acting illegally
This opposition in content signals that the cum here is concessive, meaning:
‘Although he was praiseworthy in his childhood and obeyed his teachers…’
Conclusion:
Yes, ‘cum in pueritia sua laudabilis esset et… obtemperaret’ is a concessive cum-clause, because it sets up a contrast with the main clause, and both verbs are in the imperfect subjunctive, which supports the concessive use.
As it can be seen, initially ChatGPT’s analysis of the Latin sentence ‘cum in pueritia sua laudabilis esset et…obtemperaret’ was incorrect, since it recognised it as a temporal clause, but after more detailed questions, it finally recognised correctly as a concessive clause.
Then we moved on to a lexical analysis. At the beginning, students worked independently to search the Latin vocabulary for the words related to the semantic field of ‘potestas’ and ‘educatio’ and they were invited to use GAI to find and compare their results with the ones of GAI. Then, they were divided into groups of four people and used the list of terms created with the support of GAI to write short sentences and ask GAI to correct them.
From the activity of this module, we can draw some interesting considerations:
-
• The students engaged with interest and actively used GAI as a support.
-
• Comparing their work of analysis and translation to GAI work, they were able to find that GAI does not always analyse or translate accurately (as seen, for example, with the first sentence of the Latin text above), so they must use a more critical approach with GAI.
-
• Now they know what the hallucinations are and their consequences.
-
• They discovered that it is important to use a conversational approach to arrive at the right results and to receive a more profound explanation about how to translate correctly.
Home assignment: To reinforce and expand upon the knowledge gained in this lesson, students were assigned homework: the translation of another text containing vocabulary related to education and most of the syntactic rules present in the class text. The assignment was the following:
-
• Analyse and translate the attached Latin text: ‘De disciplina in scholis’.
-
• Summarise the text.
-
• After completing the analysis, translation, and summary of the text, ask the GAI to perform the same tasks. Then, compare your own work with the GAI’s.
-
• Identify any parts where the GAI’s translation differs from yours and explain where the mistake lies and why, using the GAI’s help.
-
• (Use correct prompts to ask the GAI for explanations, referring to Latin grammar rules. Example prompt: ‘Can you explain why the GAI translated this word/phrase this way? Is it correct according to Latin grammar rules?’)
-
• Review the morphosyntactic rules that you got wrong in your translation using your grammar textbook. You can also use the GAI to ask for further clarification of these rules.
-
• (Example prompt: ‘Explain the use of the subjunctive in this sentence’ or ‘Why is this verb in the ablative absolute?’)
-
• Submit the complete work, including your prompts, using the ‘Share’ button on ChatGPT or by copying and pasting everything into a document.
De disciplina in scholis
In scholis Romanis magna semper diligentia laboratur. Discipuli, qui litteras et artes discere volunt, a magistris severis saepe monentur ut studium augeant. Praeceptores, quorum auctoritas summa est, discipulos hortantur ne umquam pigri sint, sed ut cotidie labores maiores suscipiant. Si quis discipulus tardus aut neglegens est, saepe increpatur, non odio, sed amore, ut melius discat. Omnes enim sciunt nullam virtutem sine labore exerceri posse. Discipuli diligentes praemia accipiunt, quae non pecunia, sed laude et honore constant. Magna laus datur eis, qui non solum docti, sed etiam moribus boni sunt.
Fourth module: stylistic analysis and creative use of GAI
This module was more focused on stylistic analysis and creative use of GAI.
The activity was collaborative and based on two passages: one from Plautus’ Menaechmi, Actus I, ii (Rubino Reference Rubino2008), and the other from Cicero’s De Oratore, Book I, section 22 (Norcio Reference Norcio1970), both accompanied by a parallel translation.
Students were initially divided into several groups of three or four students.
Each group was asked to identify the stylistic features of both texts, supported by the teacher, and to write down the differences observed between the two from a stylistic point of view.
After logging into ChatGPT or a similar software, students asked the GAI to perform the same task: to identify recurring colloquial expressions in the Plautine text and rhetorical or formal expressions typical of Cicero’s oratory. The GAI-generated results were then compared with a sample of student responses.
Then, I asked each group of students to produce creatively Plautine-style dialogues between fictional characters (e.g., an adulescens and a senex, or a pater and a matrona), using expressions they had learned. These dialogues were based on imagined situations from daily life in ancient Rome (e.g., at the market, in the forum, or at a banquet).
An example is the following:
The Cunning Wife and the Jealous Husband
Giancarlotto: Tell me, my wife, what did you do last night?
Giuseppina: I went to my friend Ingrid’s home.
Giancarlotto: By Jove, woman of loose morals, how dare you lie to me?
Giuseppina: My dear husband, would I, a woman devoted to you, ever be capable of hiding the truth from you and telling you silly lies?
Giancarlotto: I don’t doubt your fidelity, but you must not disobey me.
Giuseppina: I’m not disobeying you, we were out of onions for dinner, so I went to Ingrid’s to get some.
Giancarlotto: Oh, Giuseppina, I am so grateful to have a devoted wife like you…Yet I could have sworn I saw Fabio leaving here yesterday morning.
Giuseppina: My love, surely you’re not going to insult my friend Filomena by saying she looks like a real man?
Giancarlotto: No, you’re right; my eyes must have played tricks on me for a moment, and my mind wandered to dark places.
Then, students performed their dialogues orally, like short theatrical scenes, in front of the class.
Every student subsequently voted for the dialogue that was best executed and most faithful to Plautus’ style (the one above won).
Finally, we asked ChatGPT to give its opinion about the best theatrical dialogue written by students:
ChatGPT response: Yes, the scene captures the comedic tension and clever reversals typical of Plautus, especially in its premise and character dynamic. With a bit more farce, verbal flourish, or exaggerated persona, it could slide easily onto a Roman stage.
ChatGPT’s evaluation was pretty good and underscored the students’ comprehension of Plautus’ theatrical style and characters.
A concluding discussion took place that included a guided reflection on the differences observed between completing the activity traditionally and with the support of ChatGPT.
The students participated with interest in all activities of this module, and they perceived that the mixture of traditional learning with GAI was more effective.
Fifth module: final assessment
In this last module, students were given a Latin-to-Italian translation to complete within an hour and a half.
The text was of the same level as the one proposed in the second lesson, by the same author, in the same style, with the same rules, and many words belonging to the same semantic fields (power, education).
The module final activity was a homework assignment consisting of reviewing the translation completed at home with the support of the GAI, along with the completion of the final questionnaire on acquired knowledge and students’ perceptions of the GAI and the project evaluation survey.
The goal was to assess how the use of GAI had influenced students’ translation outcomes, as well as how it had affected their study methods, their approach to Latin, and possible developments going forward.
Comparing the results of the two assignments, the first completed at the second meeting and the second at the last meeting, showed that out of 32 completed submissions, the differences between the first and second were clear: 11 showed evident improvements, 10 had no clear improvements, and 9 were slightly worse.
While the GAI intervention didn’t significantly impact the translation skills of all students (nor should this be expected given the project’s limited duration), they all reported on the final questionnaire that they all changed their approach in using GAI, making them more proficient and aware in using GAI to support their learning, with a better understanding of its limitations and the importance of ethical use.
Findings and observations
Overall, the project yielded the following insights:
Understanding of Basic Concepts of GAI: Most students reported having understood the concepts either partially or fully. Notably, no student stated that they did not understand them, suggesting that the level of difficulty was appropriate for their age group while still leaving room for deeper exploration or clarification on specific topics.
Perceived Usefulness: The project was perceived as useful by all students. Many noted that, without this experience, they would not have received similar training, underscoring the innovative value of the initiative. This highlights a key takeaway: the project successfully addressed a perceived educational gap.
Ideal Applications of GAI: Student responses indicate a recognition of various intelligent uses of GAI, particularly the potential to receive personalised, guided responses and support in consolidating concepts or translating and analysing texts. This suggests that students are beginning to view GAI not merely as a translation tool, but as a study partner, a significant paradigm shift worth reinforcing (Chiu and Rospigliosi Reference Chiu and Rospigliosi2025).
Improvement in Latin Studies: Almost all students reported that using GAI helped them improve their understanding, analysis, or translation of Latin texts. This confirms the practical effectiveness of the proposed approach.
Interest in Continuing: Responses to this question were more varied: some students were enthusiastic about continuing to explore GAI in Latin; others were uncertain; very few were uninterested. The uncertainty may stem from a perception of GAI as a tool ‘foreign’ to the humanities. Strengthening the connection between GAI and classical activities (text commentary, composition, and language games) could help boost motivation.
Student Feedback: Students generally rated the explanations as very clear or fairly clear. The materials used during the project were considered very useful or adequate, and most students judged the duration of the project activities to be appropriate.
Conclusions
Our ‘AI and Latin’ project invites a series of meaningful considerations on the evolving role of education in the age of GAI. In an era increasingly shaped by the pervasive use of GAI across various sectors, it is imperative that schools adopt a more open and forward-looking stance, embracing emerging technologies rather than resisting them. In this context, it is no longer tenable to rely exclusively on traditional pedagogical approaches. Educators must not only become proficient in new technological tools but also develop a reflective understanding of how these innovations can transform their teaching practices (Lee et al. Reference Lee, Arnold, Srivastava, Plastow, Strelan, Ploeckl, Lekkas and Palmer2024).
These reflections emerge from concrete evidence observed during the implementation of the project. First and foremost, the project proved to be both remarkable and impactful: it was highly appreciated by students, stimulated genuine interest, and significantly contributed to the development of their linguistic skills. The meaningful response from participants confirms that the integration of AI into the study of the humanities, Latin in particular, is not only feasible but also welcomed, provided that it is framed within a didactic model that is accessible, critical, and creatively designed.
A fundamental point must be emphasised: these tools are not intended to replace teachers, but to support and enhance their pedagogical strategies (Chiu and Rospigliosi Reference Chiu and Rospigliosi2025). The activities proposed throughout the project exemplify this principle. GAI tools, particularly conversational agents, demonstrate a remarkable versatility, offering dynamic, interactive, and student-centred learning experiences. Their capacity to simulate dialogue, provide immediate feedback, and scaffold learning processes makes them especially valuable in the study of ancient languages, where active engagement and contextual understanding are key to mastery.
Furthermore, the integration of GAI in the Latin classroom aligns with recent advances in cognitive science and educational neuroscience, which underscore the importance of multimodal, personalised, and inquiry-based learning environments. By incorporating GAI into Latin instruction, we can foster a richer and more meaningful connection between students and the classical world, one that transcends rote memorisation and grammatical drills to emphasise critical thinking, cultural reflection, and linguistic creativity.
In conclusion, the ‘AI and Latin’ project demonstrates that the future of classical education lies in a hybrid model where tradition and innovation converge. The integration of GAI into Latin pedagogy offers a powerful opportunity to revitalise the discipline, making it more relevant and engaging for today’s learners. This is not merely a technical enhancement but a cultural shift, a new paradigm in teaching and learning that invites both educators and students to reimagine the boundaries of the classical canon.
Open access
