The Impact of Ancient Greek Prompts on Artificial Intelligence Image Generation: A New Educational Paradigm

Abstract

Background/Objectives: This article explores the use of Ancient Greek as a prompt language in DALL·E 3, an Artificial Intelligence software for image generation. The research investigates three dimensions of Artificial Intelligence’s ability: (a) the sense and visualization of the concept of distance, (b) the mixing of representational as well as mythic contents, and (c) the visualization of emotions. More specifically, the research not only investigates AI’s potentialities in processing and representing Ancient Greek texts but also attempts to assess its interpretative boundaries. The key question is whether AI can faithfully represent the underlying conceptual and narrative structures of ancient literature or whether its representations are superficial and constrained by algorithmic procedures. Methods: This is a mixed-methods experimental research design examining whether a specified Artificial Intelligence software can sense, understand, and graphically represent linguistic and conceptual structures in the Ancient Greek language. Results: The study highlights Artificial Intelligence’s possibility in classical language education as well as digital humanities regarding linguistic complexity versus AI’s power in interpretation. More specifically, the research not only investigates AI’s potentialities in processing and representing Ancient Greek texts but also attempts to assess its interpretative boundaries. The key question is whether AI can faithfully represent the underlying conceptual and narrative structures of ancient literature or whether its representations are superficial and constrained by algorithmic procedures. The study highlights Artificial Intelligence’s possibility in classical language education as well as digital humanities regarding linguistic complexity versus AI’s power in interpretation. Conclusions: The research is a step toward a more extensive discussion on Artificial Intelligence in historical linguistics, digital pedagogy, as well as aesthetic representation by Artificial Intelligence environments.

1. Introduction

This research seeks to bridge the fields of linguistics and Artificial Intelligence, using texts written in the Ancient Greek language, with an AI-based image generation tool. The text that has been chosen is included in the curriculum of Greek public education. The novelty of the project establishes connections between AI and Ancient Greek—a language with a highly complex syntactic structure that is no longer in active use. It aims to seek the results of using such a pioneer method while combining AI and new teaching opportunities. AI tools produce images based on prompts, which serve as instructions that guide the content creation. This process involves interaction between the user-teacher and AI. The subject of this research is the way AI interacts with a thousand-year-old non-speaking language, and investigates such aspects as the degree of perceptiveness, the accuracy of visualization, the interpretation of the content, and the role of the teacher.

1.1. The Architectural Structure of the Ancient Greek Language

Ancient Greek is more than a means of expression; it is a fundamental component of thought formation and Greek identity [1] (pp. 17–19). The linguistic architecture is highly complex and is structured in varying levels that determine its functionality. There is a direct relation between thinking and language because language is a means of thinking construction as well as concept formation. The sentence (etymologically from the verb “προτείνω”, which means proposing) is a single linguistic construction unit, as it transmits complete thoughts as well as makes conversation construction possible. The verb “to propose” itself originally served to describe the activity: “to stretch towards” with emphasis on the metaphorical notion of linguistic conversation as a matter of information as well as transmitting ideas [2] (p. 118). The sentence is the central unit of language and forms the basis of language structure, as it expresses an integrated thought. Syntax is based on the combination of the subject (the name or phrase that indicates who or what acts) and the predicate (the verb or phrase that expresses what is happening to the subject), this relationship shapes communication and allows the expression of complex concepts. The syntactical structure of Ancient Greek, which is one of the synthetic languages, exhibits form changes (external as well as internal) to convey syntactical as well as semantic relationships and makes use of the following:

a.

Syntactic complexity: There are often instances where the subject and object are connected in different ways than in modern Greek or other languages.

b.

Conjugation of nouns: Cases, such as nominative, genitive, dative, accusative, and vocative determine the syntactic role of words within the sentence.

c.

Conjugation of verbs: time, tense, and mood change the form of verbs.

d.

Composition and inflection of words, differ from modern Greek, which is more analytical and uses more circumlocutions.

e.

Ancient Greek exhibits strong morphological complexity, with the use of imperative prefixes and suffixes (e.g., ἀπο-, ὑπερ-, ὁλο-) that reinforced the meaning of words [3] (pp. 226–230).

The complexity in its form makes its study in high schools problematic for students because it is not easy for them to relate Ancient Greek with their daily existence, as its syntactical as well as grammatical analysis tends to remove its live quality. Despite centuries having gone by, a great deal of the old Greek language is still in the modern Greek language. The preservation of its vast inflectional morphology, its use of synthetic words, as well as its preservation as a connector with its philosophic as well as rhetorical basis, is a testament to its enduring quality. However, the change in pronunciation and the simplification of grammar are characteristics of the evolution of any natural language [2] (pp. 33–34). Ancient Greek is not only a system of communication but also a way of thinking that has deeply influenced Western civilization. Its complex form, its differentiation of pronunciation, and its syntactic complexity make it a challenging but fascinating language, which continues to be an object of study and admiration.

According to a 2023 study conducted by a research team at the University of Cambridge on the capabilities of ChatGPT 3.5 in teaching ancient languages, focusing on Latin, Ancient Greek, and Classical Sanskrit, it appears that among these three ancient languages, Ancient Greek is the biggest challenge for ChatGPT 3.5, which highlights the distinctiveness of the Ancient Greek language. Regarding Latin, ChatGPT is very capable of dealing with grammar and English-to-Latin translation and is capable of producing largely correct Latin compositions with minor errors. However, it struggles with full morphological analysis as compared to specialist tools like Whitaker’s Words. Its English-to-Latin translation is functional but less reliable, tending to provide unnatural phraseology. When it comes to Classical Sanskrit, ChatGPT is capable of analyzing grammar well but tends to provide erroneous or ambiguous examples. Its Sanskrit-to-English translation is fairly accurate, though with some semantic errors, while its English-to-Sanskrit translation is less correct, tending to lean towards free interpretation. While it is capable of composing text in Sanskrit with reasonable accuracy, verification is needed. In contrast, ChatGPT encounters significant difficulties with Ancient Greek. It tends to introduce features of modern Greek, creating hybrid sentences that do not conform to ancient linguistic norms. Its translation from Ancient Greek to English is inconsistent, often containing grammatical and syntactical errors. Furthermore, its English-to-Ancient Greek translation is practically unusable, as the output does not reflect authentic Ancient Greek [4] (p. 158).

1.2. DALL·E Image Generator-Technological Features

Here, in this research work, we will be making use of the OpenAI-developed DALL·E image generator, which is a state-of-the-art image generator. Specifically, we used version ChatGPT-4 Turbo for our experimental research on 26–27 February 2025. Released in October 2023, DALL·E improved on the functionality of its predecessor models by improving both caption quality as well as picture quality. Improving on its predecessor models, it enhanced on input text-pictures output combination towards producing images highly like input descriptions. The users can avail of DALL·E through ChatGPT Plus as well as Enterprise, which can be utilized by users to produce new images straight from conversation prompts [5] (pp. 7–8).

The underlying operation in this device is a procedure that employs natural language processing (NLP), in addition to deep learning models as well as computer vision models to create high-quality images. After receiving a prompt from a user, AI must interpret as well as analyze human language. To start with, words are tokenized into a structured format that can be interpreted by AI by DALL·E. The CLIP (Contrastive Language-Image Pretraining), a pre-trained AI that is familiar with interpreting descriptions in words as well as images, is engaged by the system. The CLIP ensures that AI is interpreting the meaning of words. The prompt is interpreted by DALL·E into a representation, a mathematical encoding in terms of images’ visual attributes. The model aggregates features that have been requested from a massive training database. The images that are output from the device are iteratively refined until a realistic capture is obtained from a prompt. The output can be iteratively improved by a user at the final stage. The procedure is a stepwise procedure that employs NLP, CLIP encoding, diffusion modeling, as well as iterative refinement to translate words into a convincing picture. The AI model is continuously improved through reinforcement learning and user feedback [6] (pp. 3–8). Machine learning has influenced the effectiveness of AI-generated content.

When referring to prompting in Ancient Greek, machine learning models such as OpenAI’s DALL·E leverage linguistic datasets to interpret and generate historically accurate visuals. The syntactical complexity of Ancient Greek with its highly intertwined syntactical relationships presents a unique challenge in AI processing. Recent studies have identified that AI-optimized developments in text-to-picture can add quality to output by extracting information from multimodal sources of data that are old texts or paintings. In addition, AI models are enhanced in interpreting historical culture as well as literary prompts through reinforcement learning methods that add pedagogic possibilities in AI-generated images [7] (pp. 112–114).

Nevertheless, at this point a clarification is necessary. DALL-E is designed to automatically translate image descriptions provided in non-English languages into English before generating the corresponding image. This process occurs because most large multimodal models, including DALL-E, are predominantly trained on English-language data. Consequently, translation enhances the model’s comprehension of the request before the image generation phase, during which techniques such as CLIP are employed to establish associations between textual descriptions and visual outputs.

1.3. Human-Centered Approach to the Use of GenAI in Education

The use of GenAI tools in the field of education not only requires knowledge and skills but also demands an ethical and moral balance with human creativity. Furthermore, the educator’s role should not be displaced to ensure a human-centered approach to the use of AI in the educational framework. This need is also foreseen by the United Nations Educational, Scientific and Cultural Organization (UNESCO), and the report published in 2024 on the official website of the organization entitled “Guidance for generative AI in education and research” provides policymakers with guidance towards a human-centered approach to the use of GenAI. Recognizing the fact that the available GenAI tools are being updated at a faster pace than national regulations can keep up with, UNESCO seeks to support countries in their efforts to take steps to design long-term policies to ensure a human-centered vision in the use of new technologies. It evaluates potential risks and stresses the need for validation of GenAI systems by academic and educational institutions as to their ethical and pedagogical relevance. Summarizing a series of recommendations proposed by UNESCO to policymakers and educational institutions, it is observed that emphasis is placed on needs such as ensuring social inclusion and educational equity, with particular attention given to marginalized groups, minorities, and people with disabilities. It particularly stresses the need for regulation and ethical guidance and calls on educational institutions and governments to develop policies on the ethical use of GenAI and to define the limits of the use and protection of personal data. Also, to strengthen the skills of teachers, the United Nations Educational, Scientific and Cultural Organization (UNESCO) proposes training programs to enable them to integrate GenAI tools into their teaching in a way that supports students’ creativity. It does not overlook the need to preserve cultural and linguistic diversity and therefore emphasizes GenAI systems to adapt to local cultural conditions by promoting the use of a variety of languages in addition to the dominant Western languages. Finally, it raises the issue of the validation of AI platforms following evaluation by academic and regulatory bodies, to ensure their reliability and suitability for the educational process. It is of course not only the ethical responsibility of government regulatory bodies and institutional users of these tools, such as universities and research institutions, but it should also be the responsibility of the GenAI tool providers themselves, who are obliged to develop technologies that comply with regulatory standards, and individual users, such as teachers and students, should use these tools responsibly, understanding their limitations and thinking critically, avoiding uncritical acceptance of information [8] (pp. 30–35).

Additionally, the European Commission highlights a framework that ensures that technology in education is used ethically, inclusively, and responsibly, by managing, protecting, and sharing digital resources and by taking proactive measures to protect students’ physical, psychological, and social well-being while using digital tools. This includes fostering critical thinking, ensuring accessibility and inclusion, and addressing concerns such as screen time, digital addiction, and ethical dilemmas related to AI and automated systems [9] (pp.70–75). It is on these ethical and moral grounds that Ancient Greek is placed as a prompting language on a picture-generating device with AI as a pedagogic instrument. This approach reflects a respectful stance toward linguistic diversity, the preservation of a historical language, and its integration into contemporary technological developments, thereby fostering local distinctiveness and promoting Greek cultural identity.

2. Materials and Methods

2.1. Text Material

The educational paradigm of the present research is developed by using a selected text from an Ancient Greek source. It is an excerpt from a fantasy novel by the second-century satirist and rhetorician Lucian. He was born in Asia, and his native language was probably Syriac. However, all of his extant works are written entirely in Ancient Greek, primarily in the Attic Greek dialect. Lucian lived from 125 to 180 AD [10]. The novel is Lucian’s best-known work and has the controversial title “True Story” (in Greek “Aληθής Ιστορία”) although it is, in fact, a fictional novel. The novel is a satire of outlandish tales that had been reported in ancient sources, particularly those that presented fantastic or mythical events as if they were true [10]. The imagery described in it combines real-life iconic references (e.g., sea, island, ship, people) with fictional representations (e.g., creatures walking on the sea with legs made of cork). It also describes visual emotional shadings (e.g., people looking on in wonder) and numerical determinations (e.g., distances and sizes). It is considered the first known science fiction text [11] (p. 28); therefore, it is an excellent example of how a literary description full of detailed and multi-level images can be a case study in the process of prompting.

To clarify the rationale behind the choice of this specific passage, we would like to explain the reasons. First of all, this specific text was chosen because it is part of the mandatory curriculum taught in all public and private schools in Greece to all first-year middle school students, who are encountering Ancient Greek for the first time. Through this selection, we make a realistic pedagogical proposal that is in direct harmony with the country’s educational reality. The second reason is the thematic character of this passage, with the themes of distance, representational imagination, and emotional expression—the very domains of special significance to our research. Finally, this decision also relied on the grammatical phenomena of the passage, i.e., on the participles: διαθέοντας, προσεοικότας, ὁρῶντες, βαπτιζομένους, ὑπερέχοντας, and ὁδοιποροῦντας. One of the specificities of Ancient Greek as a synthetic language is the availability of participles, as each participle indicates several aspects, i.e., tense, adverbial meaning, person, and case. In both Modern Greek and English—the language spoken by ChatGPT—each participle is rendered as a full adverbial clause. This is evidence that Ancient Greek was a highly complex and dense language, which presents huge challenges to modern students, who speak an analytical language these days and struggle to grasp the condensed meanings of Ancient Greek verbal forms.

Here is the entire ancient text: “Πλέομεν ὅσον τριακοσίους σταδίους καὶ νήσῳ μικρᾷ καὶ ἐρήμῃ προσφερόμεθα. Μείναντες δὲ ἡμέρας ἐν τῇ νήσῳ πέντε, τῇ ἕκτῃ ἐξορμῶμεν καὶ τῇ ὀγδόῃ καθορῶμεν ἀνθρώπους πολλοὺς ἐπὶ τοῦ πελάγους διαθέοντας, ἅπαντα ἡμῖν προσεοικότας καὶ τὰ σώματα καὶ τὰ μεγέθη, πλὴν τῶν ποδῶν μόνων· ταῦτα γὰρ φέλλινα ἔχουσιν· ἀφ’ οὗ δή, οἶμαι, καὶ καλοῦνται Φελλόποδες. Θαυμάζομεν οὖν ὁρῶντες οὐ βαπτιζομένους, ἀλλὰ ὑπερέχοντας τῶν κυμάτων καὶ ἀδεῶς ὁδοιποροῦντας. Oἱ δὲ καὶ προσέρχονται καὶ ἀσπάζονται ἡμᾶς ἑλληνικῇ φωνῇ λέγουσί τε εἰς Φελλὼ τὴν αὑτῶν πατρίδα ἐπείγεσθαι. Μέχρι μὲν οὖν τινος συνοδοιποροῦσι ἡμῖν παραθέοντες, εἶτα ἀποτρεπόμενοι τῆς ὁδοῦ βαδίζουσιν εὔπλοιαν ἡμῖν ἐπευχόμενοι” [12].

2.1.1. Prompt Dataset

Three (3) portions from the text have been selected with care to be utilized in research, based on the syntactical complexity of the text, varied linguistic structures, iconographic descriptions both of fictional as well as real subjects, as well as variety in meaning, to investigate perceptiveness ability of the image generation tool, DALL·E, a shown in

Table 1. Text extracts for research purposes.

	Greek Text	Translation	Purpose
1st PART distance	Πλέομεν ὅσον τριακοσίους σταδίους…	We sail about three hundred stages…	Research purpose: to determine whether the AI tool can understand and capture the concept of distance and whether it can perceive the unit of measurement “stadia”, which corresponds to approximately 182 m
2nd PART fiction	…καθορῶμεν ἀνθρώπους πολλοὺς ἐπὶ τοῦ πελάγους διαθέοντας, ἅπαντα ἡμῖν προσεοικότας καὶ τὰ σώματα καὶ τὰ μεγέθη, πλὴν τῶν ποδῶν μόνων· ταῦτα γὰρ φέλλινα ἔχουσιν·…	…we distinguish many people running here and there on the sea, who resembled us in all things, both in body and stature. excepting only the legs, because these were made of cork…	Research purpose: to determine whether the AI tool can reproduce an instruction that combines real and fictional representations
3rd PART emotion	…Θαυμάζομεν οὖν ὁρῶντες οὐ βαπτιζομένους, ἀλλὰ ὑπερέχοντας τῶν κυμάτων καὶ ἀδεῶς ὁδοιποροῦντας…	… So, we are surprised to see them not sinking, but staying on the waves and walking without fear…	Research purpose: to determine whether the AI tool can perceive the described emotions and render realistic representations

:

2.1.2. Prompting Design

This study employs a mixed-method experimental design, combining quantitative and qualitative analyses, to assess the effectiveness of Ancient Greek prompts in DALL·E for educational purposes. The methodology incorporates natural language processing (NLP), AI-based production, as well as pedagogic investigation to determine AI-produced images’ accuracy as well as relevance in terms of linguistic input in a non-spoken, old language. To clearly instruct the image generation tool, DALL·E, I first introduce explanatory instructions in the modern Greek language informing the tool that I am going to communicate by using the Ancient Greek language, and in specific Lucian’s text of the official curriculum used in public school textbooks for teaching Ancient Greek. I received a positive response and issued the first instruction: “Πλέομεν ὅσον τριακοσίους σταδίους…”, meaning “We sail about three hundred stages…” a few seconds later Figure 1 is rendered.

To examine whether the analysis of the initial prompt was random or the same prompt renders similar results, I issued the same prompt twice. So, the results of the same prompt “Πλέομεν ὅσον τριακοσίους σταδίους…” are in Figure 2 and Figure 3.

Afterward, I issued the second instruction “…καθορῶμεν ἀνθρώπους πολλοὺς ἐπὶ τοῦ πελάγους διαθέοντας, ἅπαντα ἡμῖν προσεοικότας καὶ τὰ σώματα καὶ τὰ μεγέθη, πλὴν τῶν ποδῶν μόνων· ταῦτα γὰρ φέλλινα ἔχουσιν·…” meaning “…we distinguish many people running here and there on the sea, who resembled us in all things, both in body and stature excepting only the legs; because these were made of cork…” and the second image is produced as Figure 4:

Since the initial image result did not meet my expectations, I issued a directive requesting the generation of a new image. The revised prompt specifies that the imaginary creatures should be smaller in size, appear more friendly, wear clothing, and have hair, ensuring a more refined and visually appealing outcome. The refined image follows the given instructions by incorporating the specified modifications, as Figure 5 and Figure 6:

I issue the final instruction “…Θαυμάζομεν οὖν ὁρῶντες οὐ βαπτιζομένους, ἀλλὰ ὑπερέχοντας τῶν κυμάτων καὶ ἀδεῶς ὁδοιποροῦντας ·…” meaning “…So, we are surprised to see them not sinking, but staying on the waves and walking without fear …” and the final image is produced as Figure 7 and Figure 8:

3. Results

This chapter presents the study’s findings based on three research objectives: (1) evaluating AI’s ability to interpret distance and the Ancient Greek unit of measurement (‘stadia’), (2) assessing whether AI can generate images combining real and fictional representations, and (3) determining AI’s ability to perceive and render emotions described in Ancient Greek text. Regarding the evaluation of the quality and educative worth of images generated by AI, this is something we have been ethically concerned with every time we bring AI tools into the classroom. Research questions such as: Would it be a guarantor of the ethical nature of AI in education to have a connection between AI and philosophy, and Ancient Greek philosophy in particular? Under what conditions is it appropriate to apply AI in terms of the educator’s role, the method of delivering information, access to knowledge, and the development of critical thinking? Can algorithms, and specifically modern AI models of information handling, be programmed so that learning tools can be made into digital “philosophers”?

The answers to these questions are given by Ancient Greek philosophers who sought the truth about knowledge—such as Socrates, Plato, Aristotle, and others. They developed diverse theories, yet all placed the human being in the role of a seeker of truth and knowledge [13] (p. 4). The teacher is the key person because he or she guides the student on the path of questioning without providing all the answers, leaving space for independence and mental growth. AI can provide access to vast databases and create interactive learning experiences, but at present, it is not yet capable of empathy and consciousness, nor of guiding students toward enhanced understanding in the way that human educators can [14] (p. 5). However, it is capable of promoting critical thinking when it is employed as a tool instead of an ultimate authority on information. Because of this, we chose to evaluate the images ourselves—not simply as researchers, but primarily as educators who have first-hand, human interaction within real-world educational environments.

3.1. Evaluating Distance

By utilizing the first section of the text as a prompt, this research aims to evaluate the tool’s capability to comprehend the concept of distance within Ancient Greek texts, particularly the unit of measurement known as ‘stadia’.

• Observing the initial image produced, it is firstly found that a nautical landscape was rendered in the visual representation which was derived from the verb “Πλέομεν”, which denotes a naval voyage. The aesthetics of the image refer to an ancient type of ship with people dressed in ancient clothing. The attribution of the era is not precisely described in the prompt; therefore, it was probably calculated either because of the use of an ancient language or because of the original explanations given for the use of this text, or because of the reference to the ancient unit of measurement of “stadia”.
• Concerning the calculation of distance, although it cannot be accurately represented, the measurement seems to have given a sense of perspective to the visual representation. Two ships are depicted traveling on the sea with a distance between them, and a large piece of land is depicted in the image which gives a sense of depth and distance.
• It seems that the tool encountered difficulties in visually representing specific numerical measurements, although it generally managed to illustrate the concept of distance by placing mobile and stationary elements, such as traveling ships and an island, which accentuate the concept of distance.
• In terms of the style of the image, it is observed that the aesthetics of the sunset were given with the appropriate lighting and color palette, information that was not specified in the prompt.
• It is worth noting that the tool responded using the Ancient Greek language, explaining details of the image it created it mentioned details of the text that were not given in the command; therefore, it is concluded that it was able to retrieve the text from the dataset.

3.2. Evaluating Real and Fictional Representations

The second part of the text focused on the AI’s ability to combine real and fictional representations within the same visual output. The prompt, “…καθορῶμεν ἀνθρώπους πολλοὺς ἐπὶ τοῦ πελάγους διαθέοντας… ταῦτα γὰρ φέλλινα ἔχουσιν·” (“…we distinguish many people running here and there on the sea, who resembled us in all things… except their legs, which were made of cork.”) tested whether DALL·E could realistically integrate mythological elements into a rather natural setting. In this case, in addition to the initial order, a second supplementary order was given to improve the result.

• As in the previous case, the natural landscape was accurately rendered by representing an ancient type of ship with people dressed in a way that refers to our perception of antiquity. It also seems to have correctly understood the verb in the sentence “καθορῶμεν”, which suggests that the people were looking at and observing the imaginary creatures because the representation of the human figures gives the sense that they are looking with curiosity and the posture of their bodies indicates that something outside the ship has caught their attention. It also correctly perceived the meaning of the participle “διαθέοντας”, meaning they were walking on the surface of the sea.
• In depicting the imaginary creatures that were described as having legs made of cork but which otherwise resembled humans, it failed to accurately represent the description of these imaginary creatures. Although they are successfully shown to walk on water, they are represented as naked and hairless with a physique much larger than humans. Also, an eerie feeling was given to the way the creatures moved because they were depicted as infinite and occupied the entire surface of the sea.
• A second instruction was then given asking the tool to improve the visual representation of the imaginary creatures. Specifically, it was asked to render them more friendly, with clothes and hair, and a smaller number of them. After this instruction, which was very clear, the tool was able to depict the imaginary creatures with the specific characteristics requested. It, however, failed to retain the original instruction, which stated that the creatures had legs made of cork and showed them to be entirely made of cork and they are depicted as having six toes on their feet.
• In terms of style, the representation once again uses a color palette that matches the previous images produced and gives the impression of a sunset, giving a peaceful and dreamy sense.
• It is worth noting that also in this case communication between the user and the tool is conducted in the Ancient Greek language.

3.3. Evaluating the Perception of Emotions

The final part of the text assessed whether DALL·E could visually represent emotions described in Ancient Greek. The prompt “Θαυμάζομεν οὖν ὁρῶντες…” (“So, we are surprised to see them not sinking but staying on the waves and walking without fear.”) tested the AI’s ability to convey expressions of wonder and disbelief on people and lack of fear on fictional creatures.

• The result of the visual representation did not emphasize people’s feelings of wonder and admiration, which are mainly expressed through facial expressions. To render admiration, it used people’s bodies by giving movement to the hands. It depicted the people as in previous images pointing to the imaginary creatures with their hands and looking at one of them.
• The absence of fear in the imaginary creature was rendered successfully through the calmness on his face and the smooth movement of his body.
• The AI depicted people similarly to previous images, showing them gesturing towards an imaginary creature, which served as the focal point of their admiration.
• The style, colors, and aesthetics remained consistent with earlier generated images, reinforcing artistic coherence across outputs.
• While the AI successfully captured the overall scene of astonishment, the lack of distinct facial expressions indicated that DALL·E prioritizes body language over nuanced facial details when interpreting emotional cues in Ancient Greek texts.

3.4. Overall Observations on Results

Considering the previous research objectives overall, the study presents insightful observations regarding DALL·E’s capability in processing Ancient Greek prompts. The findings reveal both the strengths and limitations of AI-generated imagery in capturing spatial concepts, blending real and fictional elements, and conveying emotions.

• AI’s Understanding of Distance in Ancient Greek-Strengths and Limitations

Strengths:

• DALL·E accurately interpreted the maritime context from the verb “Πλέομεν” (“we sail”), with a demonstration of deriving effective action-based imagery.
• Precise numerical representation was challenging, but effective visualization depth and perspective in DALL·E involved putting ships at different distances in a way that established a sense of scale in space.
• The aesthetic accuracy of the scene (ancient ships, classical clothing) suggests that DALL·E was influenced by the time the language was spoken.

Limitations:

• A lack of direct numerical interpretation makes AI models struggle with Ancient Greek units of measure and require more context.
• The unprompted addition of sunset lighting suggests that AI-generated imagery may alter historical context.
• AI’s Ability to Render Real and Fictional Elements-Strengths and Limitations

Strengths:

• DALL·E effectively captured human reactions to fantastical creatures, ensuring that human figures were positioned as observing the phenomenon.
• The supplementary instruction significantly improved the AI’s adherence to user specifications, demonstrating that iterative prompting can refine outputs.

Limitations:

• Rendering imaginary and fictional figures poses challenges, particularly when they are not commonly represented in datasets, unlike real-world elements such as ships, people, and the sea, which are frequently depicted.
• AI’s Perception and Representation of Emotions-Strengths and Limitations

Strengths:

• DALL·E effectively used body posture and gestures to convey emotions, demonstrating a general awareness of human emotional responses.
• The style, colors, and artistic coherence were maintained across images, ensuring consistency throughout the generated visuals.

Limitations:

• Facial expressions were underrepresented, with emotional intensity being conveyed primarily through body language rather than nuanced facial features.

All the aforementioned strengths and limitations are summarized in Figure 9.

Figure 9 was created in draw.io (diagrams.net), a multi-platform graph drawing software developed in HTML5 and JavaScript. 3.5. Evaluating the Rendering Stability.

A direct comparison of Figure 1, Figure 2 and Figure 3 illustrates that while DALL·E is thematically coherent when generating with an identical prompt, its images vary in terms of composition, color, and style. Figure 1 and Figure 2 are well-matched, whereas Figure 3 shows recognizable deviations, suggesting that the rendering process is not deterministically strict. These differences illustrate that while AI follows the conceptual prompt cues, the image creation remains probabilistic and not fully reproducible. This raises questions about the accuracy of AI in historical and linguistic studies, where consistency is crucial. Prompt engineering and AI training refinement can be used to enhance the control of image reproducibility for future applications.

While examining Figure 4, Figure 5 and Figure 6, the same pattern of partial consistency and variation appears. While all three pictures share major thematic elements because of the requirements of the prompt, variations in composition, detail emphasis, and stylistic approach reveal that DALL·E’s rendering process is not entirely stable. Figure 4 and Figure 5 are more proximate, whereas Figure 6 has more variations, reinforcing the idea that AI-generated works are informed by probabilistic rather than exact reproducibility concerns. This works to highlight the challenge of generating precision in AI-generated historical and linguistic visualizations and the need for more advanced prompt strategies to enhance image consistency.

Figure 7 and Figure 8 demonstrate both stability and variation in AI-generated images. While both images follow the overall themes of the prompt, variation in composition, stylistic elements, and visual emphasis suggests that DALL·E’s rendering process remains subject to chance. Though core features are preserved, the variation demonstrates that the AI does not produce fully reproducible output from identical input. This also highlights the probabilistic nature of AI-generated images and the challenge of pursuing accuracy in historical and linguistic studies.

4. Discussion-Educational Reflections, Future Works, and Conclusions

For automated computer analysis, key guiding parameters must be clearly defined. The addition of AI methods brings new possibilities as well as new choices that are not originally human in conception. The question is whether, and in which ways, a machine can be seen as creative [15] (p. 68). AI models, including DALL·E, do not inherently possess creativity but act as augmented tools that enhance human creativity. In this research, in particular, that is evident in how iteratively prompting refines output, with human input to achieve desired precision in imagining Ancient Greek thoughts. The vast and growing body of digital information that is now being created and shared is similar in revolutionary power to that brought about by the printing press, perhaps foreshadowing a shift in paradigm in technical as well as creative approaches. The transition brings into question whether AI-based creativity challenges established notions about creativity as well as authorship [15] (p. 54). Respectively the use of Generative AI in education introduces catalytic changes in educational methods and highlights the need for human guidance and validation of the results produced.

Creativity blended with innovation in research is more productive in its output, as it employs interdisciplinary methods and attempts to bridge between classical academic disciplines and new technology. That is why research in its present form is structured on that basis, which attempts to show in which ways images developed with Artificial Intelligence but with prompts in old Greek form a space of collaboration between linguistic classical studies and new technology. It is seen that in portraying images from imaginary, emotional, and real subjects that are described in linguistic terms, problematic points are involved, which are eliminated by the scientific collaboration between technology and linguistic studies. GenAI development in NLP has greatly improved computational efficiency; despite that, challenges are experienced, particularly in less-resourced languages. Such languages with little digital corpora and structured data are poorly represented in AI training models, leading to incorrect or incomplete processing. Such a scenario impacts a variety of NLP applications, from machine translation to sentiment analysis, and information retrieval, in which AI is not able to understand linguistic structures as well as culture-based nuances.

The paper highlights a threat from linguistic homogenization, in which languages dominate AI-produced outputs, leading to linguistic homogeneity. To address these imbalances, hybrid models that merge classical linguistic theories as well as structured data are proposed. Such a shift is in parallel with AI-produced images with prompts from Ancient Greece, highlighting that AI data must be enriched with historically as well as linguistically varied material to boost culture-based AI development [16] (p. 335).

This lack of representation in GenAI models makes it problematic to process and understand non-dominant linguistic forms as well as non-dominant culture contexts correctly. The argument in favor of training AI models with more varied as well as more linguistically enriched data to boost AI output on underrepresented languages as well as historical sources, is supported in this research. AI-generated images having a problem with numerical concepts, fictional content, as well as emotional expression in Ancient Greek is a direct consequence of its lack of common AI training corpora. The lack of representation in Ancient Greek as a language as well as its syntactic-morphological complexity, limits AI in understanding its complexities as much as can be. The argument in favor of socially aware AI development with lesser-used languages to facilitate equal development in technologies in education as well as in culture preservation, is supported.

The findings of this work suggest that the presence in AI datasets of less widely spoken languages, as with Ancient Greek, has profound implications for classical scholarship, the digital humanities, and language preservation. The ability of AI models, such as DALL·E, to read and visualize text representations in Ancient Greek text creates a new avenue for the interpretation of ancient linguistics. As has already been demonstrated elsewhere, AI-generated image creation and NLP software cannot process ancient languages due to their morphological complexity and the lack of large-scale digital corpora) [4] (p.146). By the use of organized data and interdisciplinary approaches that combine computational modeling with linguistic analysis, AI has the promise to aid in the revival of ancient languages and make new pedagogy possible in classical scholarship. Further, projects in the area of the digital humanities may employ AI-generated imagery to recreate ancient cultural and mythic histories, helping researchers and educators visualize past environments that had previously been available only through textual analysis [7] (pp. 118–119).

Apart from classical scholarship, the work has broader implications for the uses of AI in the preservation of language and its applicability to other ancient languages. Similar to Ancient Greek, Latin has a grammatically highly inflected system that presents issues for AI-based translation and interpretation software [4] (p. 145). By adapting AI model training to accommodate less widely spoken languages, researchers can develop more culturally rich and inclusive NLP software. The European Commission has emphasized the need to maintain linguistic diversity in the online environment, challenging AI designers to ensure that minority and ancient languages are represented appropriately [9] (pp. 80–81).

The methods employed in this work—including the utilization of AI-generated pictures to interpret and visualize linguistic form—are transferable to any other ancient language to improve the linguistic competence of AI across more ancient literature. In addition, the results are compliant with the ethical framework on AI development by UNESCO, demanding AI systems that respect and uphold cultural heritage and provide equal learning opportunities [8] (pp. 30–31). This work, thus, exemplifies the promise of AI to change the discipline of digital literacy, the conservation of endangered languages, and the frontier of computational humanities.

This study’s results demonstrate that while contemporary AI systems can generate visually captivating images from Ancient Greek prompts, they are exceedingly limited in terms of interpretive accuracy and communication of abstract meaning. This study not only explores AI’s capability for linguistic and visual representation but also critically examines its limitations in capturing the underlying conceptual and narrative frameworks of ancient texts. In the process, it demonstrates the extent to which AI-driven image generation remains governed by algorithmic patterns rather than genuine interpretative logic. The results refer to the challenge AI systems face when decoding complex literary structures and suggest the need for innovation in algorithmic design. In addition, our research opens up new avenues for the application of AI in the classical languages’ pedagogy and digital humanities—not as an autonomous interpreter but as a tool to augment human analysis. By examining the intersection of linguistic complexity and AI-driven interpretation, we enter the broader discussion of AI’s application in historical linguistics, digital pedagogy, and aesthetic representation.

One of the most interesting points in our future work is that AI could accomplish more realistic elements concerning abstraction, making this option more engaging. On the other hand, the discussion of detecting emotion in AI-generated images is intriguing, but relating this to ongoing research into sentiment analysis and visual perception could give more scientific questions and paths and new research context in our future works.