Long before the Greek alphabet, people on Bronze Age Crete used Linear A to record goods, quantities and other information. Their writing survives on clay tablets from administrative settings, vessels brought to sanctuaries and objects marked with incised signs. The objects’ findspots often reveal how the inscriptions were used. The words themselves remain largely unreadable: no generally accepted translation of the Minoan language has been achieved.
Claims that AI has deciphered Linear A often confuse identifying a document’s function with reading its language. Computers can sort signs, locate repeated sequences and check how proposed readings perform across the surviving inscriptions. Those operations do not by themselves establish what a person on Crete meant by a word 3,500 years ago. Computational advances and well-publicized proposals had not changed that status by October 2026.
The evidence available to researchers has grown. In 2024, scholars published a substantial supplement to the established corpus. An inscribed ivory object from Knossos received detailed scholarly publication in 2025, followed by new linguistic investigations in 2026. These contributions have enlarged the evidence and changed some of the questions researchers can ask, even though the language has not yet been recovered.
What Linear A Was Used For
Arthur Evans first encountered Linear A alongside the related Linear B during his early twentieth-century excavations at Knossos. He distinguished the writing systems through their signs and archaeological contexts. Linear A belongs chiefly to the Middle and Late Bronze Age, conventionally about 1800 to 1450 BCE, and appears across Crete as well as at a smaller number of Aegean sites. “Minoan” is the name modern archaeologists give to the society behind the island’s great palace centres. Its inhabitants’ own name for their language remains unknown.
Linear A is a logo-syllabic writing system. Many signs seem to stand for spoken syllables, often consonant-plus-vowel combinations, while others indicate a type of commodity or another category without spelling out a word. Administrative tablets show both kinds side by side, with syllabic groups appearing among numerals and signs for counted goods. Linear A is not an alphabet in which each sign represents a letter. Ester Salgarella’s scholarly overview of Linear A sets out this mixed structure and explains the varied surfaces on which Linear A has been preserved.
Clay tablets were closely associated with administration, but they were only one setting for the script. Sealings and roundels were used while goods were recorded or moved; writing also appears on pottery, stone vessels, libation tables and occasional metal objects. Some of these were probably associated with offerings. Their audiences and purposes differed. The quantities and layout on a delivery list provide clues even when its words are unknown, whereas a short inscription on a dedicated stone vessel leaves much less context for identifying its language.
Inscriptions have been recovered at Hagia Triada, Phaistos, Knossos, Malia, Zakros and other sites. An object’s findspot matters: a tablet found near storage areas provides different evidence from a vessel placed in a sanctuary. Crete’s palatial complexes combined economic administration with important architectural and ceremonial functions. The contents of a tablet may still be unreadable, but its association with stored goods can establish that somebody was keeping an account there.
The published corpus of Linear A has grown through major editions. Between 1976 and 1985, Louis Godart and Jean-Pierre Olivier published the five-volume Recueil des inscriptions en linéaire A, known as GORILA. A supplement issued by Maurizio Del Freo and Julien Zurbach in 2024 added 107 documents and 427 signs, increasing the total covered by that edition to 1,534 inscriptions and 7,574 signs. The Italian National Research Council documents those figures. Those figures describe one edition; later finds require the overall corpus to be counted again. An unusual inscription published in 2025 shows why the corpus must continue to be updated.

Why Linear B Could Be Read and Linear A Still Cannot
Linear B offers the closest historical comparison. It shares numerous sign forms with Linear A and developed from the Aegean writing tradition that Linear A represents. Unlike Linear A, Linear B has been deciphered: it records Mycenaean Greek. Michael Ventris established in 1952 that the language of those tablets was an early form of Greek, overturning the prevailing view that they must record a different tongue. That success provides useful methods, although it does not supply Linear A’s missing vocabulary.
Ventris drew on decades of earlier research and tested his readings against Greek. Alice Kober and Emmett Bennett had already identified patterns in sign use and the organization of the writing. Ventris arranged proposed sound values in a grid, compared recurring sign groups and used likely Cretan place names to test readings. Greek gave him an independent linguistic standard against which to check the emerging decipherment. John Chadwick subsequently helped develop its interpretation. The LiBER research resource of Italy’s National Research Council documents this sequence of research and the decisive work of 1952.
Because the scripts are related, a sign found in both can sometimes be assigned a provisional phonetic value using Linear B as a guide. Scholars can thus write a conventional transcription of certain Linear A sequences. Signs assigned the values ka, u and de, for instance, can be represented as syllables. Yet the borrowed values may not perfectly reproduce the sounds of Linear A, and pronouncing a sequence conventionally tells us little about what it meant.
Being able to sound out a foreign sentence is different from understanding it. In Linear A, even the phonetic values adopted from Linear B are not equally secure. Shared shapes indicate contact between writing traditions; they cannot demonstrate that both systems represented the same language. In his 2026 analysis of Linear A sign combinations, Brent Davis argues from combinations and distributions of signs that Linear A probably encoded a language different from Mycenaean Greek. His conclusion concerns linguistic patterns, not a newly translated Minoan passage.
Scholars have found no accepted bilingual text giving the same passage in Linear A and an understood language. Nor has a convincing linguistic relationship been established between Minoan and a language that can already be read. Comparisons with Semitic, Anatolian, Etruscan and other languages have been proposed without winning general acceptance as a decipherment. Minoan’s language family is unknown. Calling it a definite language isolate would go further. An isolate is a language whose lack of demonstrable relatives has been established through adequate comparison, and the amount of securely understood Minoan is too limited for that conclusion.
The Numbers and Repeated Phrases We Can Understand
Some inscriptions nevertheless convey information through their arrangement. Tablets combine repeated sign groups, commodity symbols and numerals in patterns resembling accounts. A particular group might designate a person, location, product or administrative category; one placed after several quantities could indicate their sum. Such interpretations start from the structure of documents rather than from a reconstructed Minoan vocabulary.
The group transcribed KU-RO is an important example. It occurs where the tablet’s arithmetic supports the interpretation “total”: the preceding figures can be compared with the number beside the final entry. The arithmetic supplies independent support for this reading; resemblance to a word in another language would offer less evidence. Still, “total” is a functional interpretation. It does not establish exactly how Minoan speakers pronounced the term or used it in ordinary grammar.
Fraction signs can be studied through the quantities recorded on tablets. People recording measured goods needed symbols for parts of a unit. In a 2021 study in the Journal of Archaeological Science, Michele Corazza, Silvia Ferrara, Barbara Montecchi, Fabio Tamburini and Miguel Valério used comparisons of shapes, statistical patterns and mathematical relationships to propose values for Linear A fractions. Their proposals could be tested against quantities on the tablets. The results improved understanding of the numerical system, while leaving the scribes’ spoken words unknown.
Even the administrative tablets give an incomplete view of Minoan life. They were often compressed records written for people who already understood the products, measures and conventions involved. A present-day receipt becomes obscure if we remove the familiar names, currency and arrangements while leaving just unfamiliar abbreviations and figures. The same problem confronts historians of the Minoan tablets, which were never intended to explain their transactions to readers thousands of years later.

Inscriptions associated with offerings pose a different kind of puzzle. Recurring sequences on stone vessels are commonly called the libation formula, although each object’s ritual function must still be assessed in its excavation context. A libation involves offering liquid to a deity or another sacred recipient. Repeated signs in a fairly regular order suggest a customary expression used in connection with such an act. It could name the giver, recipient or action, or contain an invocation. None of these possibilities has yet become an established translation.
Brent Davis studied differences between versions of the formula in his study of its word order. When one group changes position but surrounding groups stay in place, the pattern may provide a clue to sentence structure. His 2026 treatment of morphology also examines recurring beginnings and endings of sign groups as possible affixes, elements attached to words. A prefix comes before the rest of the word and a suffix follows it. Such features can suggest grammatical organization even when the individual Minoan words remain untranslated.
Grammatical clues have to be interpreted cautiously. Unrelated languages may all have prefixes, suffixes or sentences that begin with verbs, so those features alone cannot establish a language family for Minoan. Nor can the existence of an offering formula identify the deity to whom it was addressed. These vessels reveal something about the practice of writing in sanctuaries, but scholars cannot yet produce a reliable translation of a Minoan prayer from them.
What Artificial Intelligence Has Actually Achieved
Different computer methods answer different questions about Linear A. One program may catalogue individual signs and compare the handwriting of different scribes. Another may count recurring sequences and test which signs tend to appear together. A machine-learning system may examine possible relationships to better-understood languages or scripts. The programs can be useful for each specific question, but a successful classification or statistical test should not be presented as a translation.
The 2019 study by Jiaming Luo, Yuan Cao and Regina Barzilay tested a computational method on previously deciphered scripts. Using linguistic constraints and a minimum-cost-flow procedure, the researchers trained a neural model to investigate correspondences between languages. They evaluated it on Ugaritic and on already deciphered Linear B. In the Linear B evaluation, they reported recovering 67.3 percent of cognates, words related through shared historical descent. That result demonstrated a computational method under test conditions. It was not a decipherment of Linear A.
A method based on related languages works best when the related language is known or can be identified with confidence. Greek supplied that support for the study of Linear B. Linear A has no established equivalent. Software may report a close fit between proposed words, but a numerical score cannot supply a historical connection that has not been demonstrated. Trying many possible languages and hypothetical sound changes also increases the chance of accidental likenesses, particularly with short words.
Other researchers compare shapes and sign repertoires across writing systems. In a 2024 paper in Information, Aaradh Nepal and Francesco Perono Cacciafoco examined similarities between Linear A signs and those in the Carian alphabet and Cypriot syllabary using feature-based methods. They also explored language comparisons based on proposed phonetic assignments. This was an investigation of possible relationships, not an independently verified reading of the script. A few similar sign shapes mean little unless the corresponding sounds and historical linguistic connections can also be established.
A review by Pratham Manoj and Francesco Perono Cacciafoco, appearing in 2025 in a volume dated 2024, considers techniques such as generative models and minimum-cost flow. The review examines the data and assumptions those approaches require, including where they can fail. It makes no claim that ordinary Minoan sentences have been recovered by an algorithm. A 2024 systematic review in Computational Linguistics likewise sets out the obstacles between finding patterns and actually deciphering a language.
Computers can also assist with the ancient handwriting itself. Signs varied between scribes and sites, so two examples may look different despite representing the same character. Shape-comparison software can help group these variants and draw attention to unusual forms. Michele Corazza’s 2024 monograph on computational methods discusses this type of work in connection with Linear A fractions and Cypro-Minoan, another related script. More reliable classification gives linguists better evidence on which to test their hypotheses, even when it does not translate a single sentence.
Evaluating decipherment and interpretation requires different kinds of evidence. Proposed values for fraction signs can be assessed through arithmetic. Grouping tablets by layout is a claim about classification. Translating a Linear A inscription into fluent English would require a consistent account of sounds, words and grammar that works on independent documents. A polished machine-generated sentence provides none of that verification by itself.
SigLA and the Value of a Reliable Digital Corpus
Computational analysis depends on reliable records of the inscriptions. Epigraphy, the study of inscribed texts, involves documenting and assessing each surviving sign. Some tablets are damaged; others have faint incisions that scholars disagree about. An unusual mark may represent one character or another. If a digital database treats every uncertain identification as established fact, software may discover apparent patterns created by modern editorial choices rather than by Minoan scribes.
To compare inscriptions and individual sign forms, Ester Salgarella and Simon Castellan developed SigLA, the Signs of Linear A, an open digital collection of inscriptions and sign forms. Its palaeographic material lets scholars compare ancient handwriting systematically. Their 2020 publication describing SigLA explains the problem it helps address. Printed editions are essential sources, but searching and comparing thousands of individual sign shapes is much easier when readings and images have been organized consistently in a digital resource.
SigLA continues to change as new material is entered. Its public log records additions during May and June 2026, including documents from more than one archaeological site. Counts published at the project’s launch describe its contents only at that date. Nor must a digital count match the total in a printed edition, since the two may count fragments, inscribed faces or whole documents by different rules. Anyone comparing figures needs to know exactly what is being counted and when.
These editorial decisions also affect machine learning. A pattern may vanish when two signs previously treated as one are separated, or when several purportedly independent examples prove to be repetitions of the same formula. Researchers need to record each item’s origin, physical condition and transcription decisions. Increased computing power cannot make unreliable source records reliable.

New Evidence From Knossos and New Research in 2026
Recent finds show how a new inscription can alter the evidence. In a study published in 2025, Athanasia Kanta, Dimitri Nakassis, Thomas G. Palaima and Massimo Perna examined an ivory object excavated in the Anetaki area at Knossos. Kanta provisionally described it as a religious sceptre. The object consists of a ring and handle recovered in fragments from a repository inside a building linked to cult activity. The scholars catalogued the inscriptions on its two parts separately as KN Zg 57 and KN Zg 58.
Writing covers several surfaces of the ivory piece and is combined with images of vessels and animals. Twelve four-legged animals appear in separate framed spaces on part of the ring. Elsewhere, vessels are pictured beside Linear A syllabic signs. The researchers suggest that this arrangement might describe parts of a ceremony or act as a reminder of ritual procedures. Their proposal rests on the images and find context, not on a deciphered inscription. The object provides longer sign sequences to analyse and an unusual example of writing integrated with visual images.
The publication also records where the ivory fragments were found and the architectural sequence of the repository. Excavation in the area covered by the report ran from 2011 through 2017. Its scholarly publication in 2025 is a separate date from both the object’s recovery and the Bronze Age period in which it was deposited. The authors note that associated finds still need further investigation before some aspects of the deposit’s history can be described more precisely.
The clay tablets from Phaistos, shown in the following image, offer a contrast. Their columns and numerical entries provide clues to the work of administrators. The ivory piece, with writing distributed among animal and vessel images, asks different questions about ritual and the way images and text were used together.

Ester Salgarella also examines archaeological context in her 2025 study of writing in Bronze Age Crete. She considers what the surfaces, settings and objects bearing inscriptions can tell us in addition to the signs themselves. A tablet, a sealing and an inscribed offering vessel may share a writing system while serving quite different purposes. Studying their contexts helps prevent the language from being treated as though it existed apart from the people using these objects.
In The Undeciphered Aegean Scripts, published in 2026, Brent Davis examines Linear A’s phonology, morphology, syntax and patterns of syllabic signs. His published chapter summary on Linear A states that comparisons between regional inscriptions strongly support a language used across Crete that differed from the Greek recorded in Linear B. The analysis does not settle that language’s family or supply translations. The distribution of signs can thus support a limited linguistic conclusion, even while the words remain unreadable.
Davis also examines the Phaistos Disc, discovered in 1908 and marked with pictorial signs stamped in a spiral. The disc is not a Linear A tablet, and the Heraklion Archaeological Museum continues to describe its text as unresolved. Nevertheless, Davis’s 2026 comparison argues from sign combinations that its language may be strongly compatible with that underlying Linear A. The argument concerns a possible linguistic relationship and does not translate the disc. The museum describes the unresolved text, while Davis proposes a relationship based on statistical patterns.
Why a Complete Translation Remains So Difficult
The corpus may contain more than 1,500 inscriptions in the 2024 supplement, yet it remains small for the purposes of decipherment. Many pieces are brief, damaged or repetitive. Twenty vessels bearing variants of one ritual formula reveal how that formula changed, but almost nothing of ordinary conversation. Administrative records repeat numbers and signs for goods. Those patterns help identify what the texts were used for while offering only a narrow sample of the language’s vocabulary.
Establishing a sound value would still be only an early step. Scholars would then need to work out which signs make up words, how word forms vary and whether proposed meanings explain inscriptions from more than one setting. If a reading requires one set of phonetic rules for a vessel and another for a tablet, that conflict needs an explanation. A theory that survives every objection merely by introducing new exceptions is difficult to test and provides weak evidence of successful decipherment.
There is no established formula setting a minimum number of Linear A signs below which decipherment is impossible and above which it becomes certain. The number of inscriptions matters, but so do their length, diversity, repetition, reliability and relationship to independent linguistic evidence. A short bilingual inscription might resolve more uncertainties than many additional accounting tablets. An archive made entirely of abbreviated numerical records could remain frustratingly opaque even if it were much larger.
A bilingual text giving equivalent information in Linear A and a known language could help identify recurring names, titles and expressions. So might firm evidence that Minoan belonged to a particular known language family, which would allow proposed sound correspondences and grammatical patterns to be tested. Neither find is assured, and even a bilingual inscription would be a starting point for analysis of the wider corpus rather than an instant solution.
Linear B provides a useful precedent because researchers can reconstruct how it was deciphered. Researchers spent decades analysing signs and improving editions before Ventris developed a reading that could be checked against Greek. For Linear A, the same discipline means making explicit predictions and then testing them on inscriptions not used to devise the theory. Linear B’s achievement offers a method of investigation, but no guarantee that every undeciphered script will eventually be read.
What to Make of AI Decipherment Claims in 2026
In June 2026, an AI-assisted claim to read Linear A circulated widely online. Tom Di Mino, an independent researcher, argued for a Semitic interpretation and said he had used software to compare signs and proposed word readings throughout the corpus. Contemporary reporting on the claim discussed a suggested lexicon and phonetic assignments, while also noting that specialist scrutiny was still required. Reports of the proposal establish that it exists; they do not validate its reading of the inscriptions.
Independent evaluation requires published sign values, transcription choices, grammatical rules and enough worked examples for other researchers to repeat the analysis. A convincing account would also predict how previously unused inscriptions should read, without changing its rules whenever a difficulty appears. Similarities to words in another language would have to survive tests against coincidence. A claim to have translated hundreds of words is insufficient without that independent assessment.
By October 2026, no proposed interpretation had become an accepted decipherment of Linear A in specialist scholarship. New software and hypotheses are useful when they produce testable questions. Even work that fails to decipher the script can reveal weaknesses in transcriptions or suggest more reliable ways of examining a future proposal.
What a Breakthrough Would Mean for Minoan History
If Linear A could be read securely, the economic history of Bronze Age Crete would become far more detailed. The tablets might give the Minoan terms for offices, places, obligations and transactions. The numerals already show that goods were counted and recorded; understanding the associated words could reveal what the categories meant and which people or institutions controlled the movement of those goods.
Sanctuary inscriptions might offer another kind of evidence. An expression on an offering table could name a god, an individual dedicant or the purpose of a gift, but none can yet be assigned securely from the writing. Historians of Minoan religion currently work largely from objects, architectural spaces, images and their archaeological associations. A well-established reading could tie those material remains to participants’ actual words, confirming some explanations of ritual while challenging others.
Images of bulls, processions and ceremonies are a familiar part of Minoan art, including representations associated with bull-leaping at Knossos. They show what artists chose to depict, but cannot on their own explain the full meaning of those acts to the people involved. Readable documents might help distinguish palace administration from public ceremonial activity or practices confined to particular communities. Until such evidence exists, the boundaries between these categories remain uncertain.
Linear A is part of a broader set of writing systems whose signs can be studied even though their languages or meanings remain uncertain. The rongorongo inscriptions of Rapa Nui present a different combination of archaeological and linguistic challenges, yet likewise show how far pattern recognition is from demonstrated translation. Methods may be compared across cases without expecting one computational solution to fit them all.

Artificial intelligence already contributes to Linear A research through handwriting classification, comparison of sign sequences, analysis of numbers and tests of linguistic hypotheses. Enlarged editions and newly published finds have improved the evidence on which those methods work. The language remains undeciphered. Whether progress comes from another excavation, renewed examination of a known object or a computational theory, a successful proposal must account consistently for what the ancient inscriptions actually say. Fluency in a modern translation is no substitute for that demonstration.









Discussion
Comments