Skip to content
Back to the question

Evidence record

Can a lost script be read?

Every sentence on the exploration belongs to at least one claim below. Each says what kind of statement it is, what it rests on, what the evidence does not establish, and how far it was checked. Nothing here is a score, and none of it is a verdict on a belief or a theory.

Last checked 24 September 2026.

  • Source checked: 38
  • A judgement, not a fact: 11
  • Partly checked: 1

If something here is wrong, misquoted, out of date or unfair, write to support@unknownoceans.com and give the claim number, which does not change.

What the question takes for granted

A question about method with a history behind it: which lost scripts have been read, what that took, and what the unread ones lack. For the Indus signs, a prior question, whether they are writing at all, is disputed between named researchers.

  • Some lost scripts have been read

    Established

    Linear B, Egyptian and Maya are the page’s cases, each with a source.

  • Every lost script records a language

    Contested

    For the Indus signs this is the dispute itself: Farmer, Sproat and Witzel argue the signs are non-linguistic, and the Rao team argues the evidence favours writing.

  • Reading a script means recovering its language, not only its sounds

    Established

    Linear A can be sounded out approximately from Linear B values and still counts as undeciphered.

How the question is labelled

Claim: Signs without a language

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: Lost scripts

Kind
Framing. Signposting. It asserts nothing a source could check.

The question

Claim: Some have been.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

Egyptian, cuneiform, Linear B and Maya are the cases the page and its sources name.

  • Supports

    Maya Hieroglyphs: A History of the Deciphermentparagraph on Old World scripts

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Decipherment of the Old World scripts usually depended on the discovery of an inscription which says the same thing in two or more scripts

  • Supports

    SigLA: The Signs of Linear A, a Palaeographical Databasesection 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Linear B was successfully deciphered as an early form of Greek in 1952

Later work looked for

A historical account; not a finding later work revises.

Claim 3 · checked 24 September 2026 · AI-assisted audit

Claim: The ones that remain unread are missing something specific,

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s summary of the sections below, where each unread script is described by what it lacks.

Claim 4 · checked 24 September 2026 · AI-assisted audit

Claim: and in one case the argument is about whether it is writing at all.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

The Indus signs. Farmer, Sproat and Witzel argue they are not writing; the Rao team argues the evidence favours writing without proving it.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 5 · checked 24 September 2026 · AI-assisted audit

Claim: Decipherment has needed two of three things: a known language, a bilingual text, a large corpus.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s generalisation from the three cases it describes (claims 17 to 20). No source states it as a rule.

Corrected

24 September 2026. Previously: “Decipherment needs a known language, a bilingual text, or a very large corpus.” The summary said any one of the three suffices, which contradicted the passage below ("Two of three things make it possible") and the cases it describes. Brought into line, and put in the past tense as a reading of cases rather than a law.

Claim 6 · checked 24 September 2026 · AI-assisted audit

Claim: The famous unread scripts each lack at least two of those,

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

Follows from the page’s account of the Indus signs (claim 25) and Linear A (claims 59 to 61).

Claim 7 · checked 24 September 2026 · AI-assisted audit

Claim: and for the Indus signs there is a serious argument that they do not encode language.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

"Serious" is the page’s word. The argument has named opponents, and the page reports them.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 8 · checked 24 September 2026 · AI-assisted audit

Claim: Egyptian hieroglyphs fell to a bilingual inscription.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Maya Hieroglyphs: A History of the Deciphermentparagraph on Old World scripts

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: such as the Rosetta Stone in which two forms of Egyptian writing are followed by Greek

What this does not establish

The source names the bilingual; it does not tell the rest of the Egyptian story.

Later work looked for

A historical account; not a finding later work revises.

Claim 9 · checked 24 September 2026 · AI-assisted audit

Claim: Linear B fell to a guess about place names that pointed to Greek,

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    The Decipherment Processpage 3 to 4, and the 1952 broadcast quoted there

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Ventris guessed that certain words on the Knossos tablets, which often appeared in headings, might be place-names ... all this pointed to a result that Ventris had never expected ... be written in Greek

Later work looked for

A historical account; not a finding later work revises.

Corrected

24 September 2026. Previously: “Linear B fell to a guess that the language was Greek, which its excavator had denied for forty years.” Greek was not the guess. The Cambridge account has Ventris guess that certain words were place names; the readings those gave pointed to Greek, which he had not expected. The opening had no source for the Evans detail; one is now cited.

Claim 10 · checked 24 September 2026 · AI-assisted audit

Claim: a language its excavator had dismissed for forty years.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

Evans excavated Knossos from 1900 and died in 1941, eleven years before the decipherment.

  • Supports

    The Decipherment Processpages 1 to 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Very few scholars thought Greek was a serious contender, largely because Evans had dismissed this possibility entirely. ... From 1900, ... at the time of his death in 1941

Later work looked for

A historical account; not a finding later work revises.

Claim 11 · checked 24 September 2026 · AI-assisted audit

Claim: What remains unread is not harder in the same way.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s framing of the sections below.

Claim 12 · checked 24 September 2026 · AI-assisted audit

Claim: It is missing different things.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

Follows from the page’s account of what each unread script lacks.

Claim 13 · checked 24 September 2026 · AI-assisted audit

What decipherment needs

Claim: What decipherment needs

Kind
Framing. Signposting. It asserts nothing a source could check.

What a decipherment needs

Claim: What a decipherment needs

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: Two of three things make it possible: knowing the underlying language, having the same text in a known script, or having enough material for statistical attack.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s own generalisation from Linear B, Egyptian and Maya (claims 17 to 20). The Penn Museum article says Old World decipherments usually depended on a bilingual.

  • Context

    Maya Hieroglyphs: A History of the Deciphermentparagraph on Old World scripts

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Decipherment of the Old World scripts usually depended on the discovery of an inscription which says the same thing in two or more scripts

Claim 16 · checked 24 September 2026 · AI-assisted audit

Claim: Linear B had a large corpus and an unknown language,

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    The Decipherment Processpages 1 to 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: produced thousands more inscriptions ... knew what the language of the inscriptions was

Later work looked for

A historical account; not a finding later work revises.

Claim 17 · checked 24 September 2026 · AI-assisted audit

Claim: and was broken in 1952 by identifying the language as an early form of Greek.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    SigLA: The Signs of Linear A, a Palaeographical Databasesection 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: Linear B was successfully deciphered as an early form of Greek in 1952

  • Supports

    The Decipherment Processpage 4

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: all this pointed to a result that Ventris had never expected

Later work looked for

A historical account; not a finding later work revises.

Claim 18 · checked 24 September 2026 · AI-assisted audit

Claim: Egyptian had a bilingual text.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Maya Hieroglyphs: A History of the Deciphermentparagraph on Old World scripts

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: such as the Rosetta Stone in which two forms of Egyptian writing are followed by Greek

Later work looked for

A historical account; not a finding later work revises.

Claim 19 · checked 24 September 2026 · AI-assisted audit

Claim: Maya glyphs had a partial key in a colonial manuscript and a living related language.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

The manuscript is Diego de Landa’s Relación de las Cosas de Yucatán, thought to have been written in 1566.

  • Supports

    Maya Hieroglyphs: A History of the Deciphermentopening and the section on Landa

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: The language has not died out: it is still the native tongue of more than a million people in Mexico and Central America. ... it is actually a partial syllabary matching the spoken sounds of the Spanish alphabet

What this does not establish

The article, from 1985, also says Landa’s key was not enough on its own, and that the script is heavily logographic.

Later work looked for

A historical account; not a finding later work revises.

Claim 20 · checked 24 September 2026 · AI-assisted audit

Why the Indus signs are different

Claim: Why the Indus signs are different

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: The corpus is about four thousand inscriptions,

Kind
Present-day figure. A count, survey or report finding, as the named source gives it.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

Both sides of the dispute give the count; Farmer, Sproat and Witzel put it at four to five thousand objects.

Later work looked for

Crossref, 2011 on, "Indus script" and "Indus inscriptions": no newer corpus count or length figure in a reviewed source; the Rao team (2010) and Farmer, Sproat and Witzel (2004) figures still stand as the ones both sides use.

Corrected

24 September 2026. Previously: “The corpus is about four thousand inscriptions, most of them five signs long, on seals and tablets.” "Most of them five signs long" misread an average as a typical length: Farmer, Sproat and Witzel give an average of 4.6 signs over 2,905 objects, many with one or two signs. "Seals and tablets" narrowed to what the Rao team says: the majority are on stamp seals.

Claim 22 · checked 24 September 2026 · AI-assisted audit

Claim: the majority on stamp seals,

Kind
Present-day figure. A count, survey or report finding, as the named source gives it.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" and "Indus inscriptions": no newer corpus count or length figure in a reviewed source; the Rao team (2010) and Farmer, Sproat and Witzel (2004) figures still stand as the ones both sides use.

Claim 23 · checked 24 September 2026 · AI-assisted audit

Claim: averaging fewer than five signs.

Kind
Present-day figure. A count, survey or report finding, as the named source gives it.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" and "Indus inscriptions": no newer corpus count or length figure in a reviewed source; the Rao team (2010) and Farmer, Sproat and Witzel (2004) figures still stand as the ones both sides use.

Claim 24 · checked 24 September 2026 · AI-assisted audit

Claim: There is no bilingual text, the language is unknown,

Kind
Observation. Something measured or seen, as the named source reports it.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" and "bilingual": no reported bilingual; nothing that changes it.

Claim 25 · checked 24 September 2026 · AI-assisted audit

Claim: and the average inscription is shorter than this sentence.

Kind
Present-day figure. A count, survey or report finding, as the named source gives it.
Review
Source checked. The cited passage was read, and it says this.

The sentence has eighteen words; the average inscription has fewer than five signs.

Later work looked for

Crossref, 2011 on, "Indus script" and "Indus inscriptions": no newer corpus count or length figure in a reviewed source; the Rao team (2010) and Farmer, Sproat and Witzel (2004) figures still stand as the ones both sides use.

Claim 26 · checked 24 September 2026 · AI-assisted audit

Claim: In 2009 a study applied a statistical measure of how ordered the sign sequences are and reported that the Indus pattern matched known writing systems and stayed far from the non-linguistic sequences it was compared with, two of which were generated artificially.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

The measure was bigram conditional entropy. The other two non-linguistic comparisons were DNA and protein sequences.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “In 2009 a study applied a statistical measure of how ordered the sign sequences are and reported that the pattern resembles known writing systems more than it resembles non-linguistic sign systems.” The paper compared no real non-linguistic sign system: its non-linguistic comparisons were two computer-generated data sets, DNA and protein. That is the centre of the objection that follows, and the page now says it.

Claim 27 · checked 24 September 2026 · AI-assisted audit

Claim: The result was contested immediately and at length, on the grounds that the measure does not distinguish language from other structured symbol systems.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

Online criticism began within days of online publication; the journal exchange followed in 2010.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 28 · checked 24 September 2026 · AI-assisted audit

Claim: Its authors replied that they had never claimed the measure proves language on its own, only that it adds weight to the script’s other language-like features.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Entropy, the Indus Script, and Language: A Reply to R. Sproatsection 3

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: We do not make such a claim; instead, we only note in Rao et al. (2009a) that our result increases the evidence for the linguistic hypothesis, when one takes into account other language-like properties of the script

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “” The page reported the objection and not the reply, though both ran in the same journal. The Rao team’s reply is now cited and stated in its own terms.

Claim 29 · checked 24 September 2026 · AI-assisted audit

Claim: The exchange ran in a computational linguistics journal,

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

Computational Linguistics, volume 36, issues 3 and 4 (2010).

  • Supports

    Entropy, the Indus Script, and Language: A Reply to R. Sproattitle, section 1 and first-page footer

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: In a recent Last Words column (Sproat 2010) ... © 2010 Association for Computational Linguistics Computational Linguistics Volume 36, Number 4

Later work looked for

A historical account; not a finding later work revises.

Corrected

24 September 2026. Previously: “The exchange, in a computational linguistics journal, is unusually direct about the reviewing that let the claim through.” "The reviewing that let the claim through" stated one side’s verdict as fact. It is Sproat’s argument, and is now attributed to him.

Claim 30 · checked 24 September 2026 · AI-assisted audit

Claim: and Sproat’s side of it is unusually direct about the reviewing he argues let the claim through.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

"Unusually direct" is the page’s judgement of the tone. That the reviewing was at fault is Sproat’s argument; the Rao team rejects it.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 31 · checked 24 September 2026 · AI-assisted audit

Claim: It did not end there.

Kind
Framing. Signposting. It asserts nothing a source could check.

Corrected

24 September 2026. Previously: “” Added with the report of the 2014 and 2015 exchange.

Claim: In 2014 Sproat compared a larger set of writing and non-linguistic corpora and reported that the two measures he found useful class the Indus signs as non-linguistic;

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    A Statistical Comparison of Written Language and Nonlinguistic Symbol Systemsabstract, page 457

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026 · Texts · Observational study: associations, not causes

    The source reads: Using a larger set of nonlinguistic and comparison linguistic corpora ... one of the measures proposed by Lee and colleagues (2010a) ... and a novel measure based on repetition turn out to be good measures ... both of these measures classify them as nonlinguistic

  • Supports

    A Statistical Comparison of Written Language and Nonlinguistic Symbol Systems§ 3, page 478

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026 · Texts · Observational study: associations, not causes

    The source reads: the statistical evidence presented here supports the nonlinguistic hypothesis for both the Indus Valley symbols and the Pictish symbols. But, as we also saw, both of our most discriminative measures, R–r and Cr, also correlate with text length

What this does not establish

One researcher’s analysis, disputed by the original authors. Sproat notes that both measures also correlate with text length, and that the non-linguistic corpora have shorter texts.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “” The currency search found the dispute continuing in Language in 2014 and 2015. The page ended in 2010 and now reports the later round, with each side’s position.

Claim 33 · checked 24 September 2026 · AI-assisted audit

Claim: the original authors replied that they stand by their findings.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Partly checked. Only part of the source could be read. What was not read is listed.

What this does not establish

Read as an abstract only: the abstract states the conclusion, and the reply’s arguments in the body were not read (no open copy was found through Unpaywall or the authors’ pages, 2026-09-24).

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 34 · checked 24 September 2026 · AI-assisted audit

Claim: Two readings of the same signs

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: It encodes language

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: Sequences show the degree of order found in writing systems

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

The Rao team’s argument. Sproat argues the same degree of order appears in non-linguistic systems (claim 41).

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 37 · checked 24 September 2026 · AI-assisted audit

Claim: Signs keep to preferred positions, as grammar would produce

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Entropy, the Indus Script, and Language: A Reply to R. Sproatsection 4

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: The script exhibits distinct language-like syntactic structure including equivalence classes of symbols with respect to positional preference, classes of symbols that function as beginners and enders

What this does not establish

Sproat’s reply lists positional preference among the properties of Mesopotamian deity symbols, which are not writing.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “Signs recur in fixed positions, as grammar would produce” The Rao team describes positional preference, not fixed positions.

Claim 38 · checked 24 September 2026 · AI-assisted audit

Claim: Short texts are normal on seals in other literate cultures

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Claim 39 · checked 24 September 2026 · AI-assisted audit

Claim: It may not

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: Statistical order also appears in non-linguistic sign systems such as the deity symbols on Mesopotamian boundary stones

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

Sproat’s test was on the Pictish measure of Lee and colleagues; the corpus was too small for the Indus measure.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “Statistical order also appears in non-linguistic sign systems such as heraldry” Heraldry was the wrong example: in the analysis Sproat reports, heraldry separated clearly from writing. His counterexample was the Mesopotamian deity symbols.

Claim 41 · checked 24 September 2026 · AI-assisted audit

Claim: No long text on any medium, and none of the traces that lost manuscripts leave in other civilisations

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

An argument from absence, made by Farmer, Sproat and Witzel. The longest inscription on a single surface has seventeen signs.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “No long texts anywhere, in a society that had them in other materials or did not need them” The old wording combined the argument with its rebuttal and read as neither. Stated now as Farmer, Sproat and Witzel make it.

Claim 42 · checked 24 September 2026 · AI-assisted audit

Claim: Near Eastern emblem systems, religious and political, look much like this

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" with "entropy", "statistical" and "nonlinguistic": Sproat 2014 (Language) and the Rao, Lee and colleagues reply of 2015 continue the dispute and are now reported on the page; the only later items are unreviewed 2026 preprints, which change nothing.

Corrected

24 September 2026. Previously: “A system of ownership marks and commodity tags would look much like this” No source on the page proposes ownership marks and commodity tags. The comparison the non-linguistic side makes is to Near Eastern emblem systems.

Claim 43 · checked 24 September 2026 · AI-assisted audit

What is missing

Claim: What is missing

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: The average Indus inscription is under five signs long.

Kind
Present-day figure. A count, survey or report finding, as the named source gives it.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2011 on, "Indus script" and "Indus inscriptions": no newer corpus count or length figure in a reviewed source; the Rao team (2010) and Farmer, Sproat and Witzel (2004) figures still stand as the ones both sides use.

Corrected

24 September 2026. Previously: “The average Indus inscription is five signs long.” The measured averages are 4.5 and 4.6 signs; five is a rounding.

Claim 45 · checked 24 September 2026 · AI-assisted audit

What machine learning changes

Claim: What machine learning changes

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: Statistical and neural methods have been applied to lost scripts, and their clearest successes are re-runs.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Deciphering Undersegmented Ancient Scripts Using Phonetic Priorsection 1 and section 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: All the known cases of lost language ... have been accomplished by human experts ... Prior work has already featured several successful instances of ancient language decipherment previously done by human experts

Later work looked for

Crossref, 2021 on, "neural decipherment", "undeciphered script" and "Linear A decipherment": no reviewed report of a machine decipherment of a script whose language is unknown; nothing that changes it.

Corrected

24 September 2026. Previously: “Statistical and neural methods have been applied to undeciphered scripts, and have succeeded where the underlying language is known or closely related to a known one.” Uncited, and "succeeded" overstated it: the successes the machine learning literature reports reproduce decipherments already made by hand. Now cited to Luo and colleagues, with the figure they report.

Claim 47 · checked 24 September 2026 · AI-assisted audit

Claim: A 2019 model matched two thirds of Linear B’s cognates to their Greek counterparts,

Kind
Experimental result. What an experiment or trial found, under its own conditions. A result in animals, cells or a simulation is not a result in people.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Neural Decipherment via Minimum-Cost Flow: From Ugaritic to Linear Babstract

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026 · Texts · Experiment

    The source reads: We also report the first automatic results in deciphering Linear B, a syllabic language related to ancient Greek, where our model correctly translates 67.3% of cognates

    67.3 per cent is within a per cent of two thirds (66.7).

What this does not establish

Measured on cognates, words with a Greek counterpart, not on the whole of Linear B.

Later work looked for

Crossref, 2021 on, "neural decipherment", "undeciphered script" and "Linear A decipherment": no reviewed report of a machine decipherment of a script whose language is unknown; nothing that changes it.

Claim 48 · checked 24 September 2026 · AI-assisted audit

Claim: a decipherment people had made in 1952,

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

A historical account; not a finding later work revises.

Claim 49 · checked 24 September 2026 · AI-assisted audit

Claim: and it was given Greek to work from.

Kind
Experimental result. What an experiment or trial found, under its own conditions. A result in animals, cells or a simulation is not a result in people.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

How the method works; not a finding later work revises.

Claim 50 · checked 24 September 2026 · AI-assisted audit

Claim: Where the related language is unknown, the methods have little to hold.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Deciphering Undersegmented Ancient Scripts Using Phonetic Priorsection 1

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: these methods are based on assumptions that are not applicable to many undeciphered scripts. The first assumption relates to the knowledge of language family of the lost language

Later work looked for

Crossref, 2021 on, "neural decipherment", "undeciphered script" and "Linear A decipherment": no reviewed report of a machine decipherment of a script whose language is unknown; nothing that changes it.

Corrected

24 September 2026. Previously: “Where the language is unknown and the corpus is small, the methods have no traction. The bottleneck is material, not cleverness, and it has been for a century.” "No traction" was uncited and absolute; the 2021 work on Iberian is a real attempt with a real, negative, result, and is now reported. "For a century" had no source and is out.

Claim 51 · checked 24 September 2026 · AI-assisted audit

Claim: Tried on Iberian in 2021, one found no strong evidence for Basque as its relative,

Kind
Experimental result. What an experiment or trial found, under its own conditions. A result in animals, cells or a simulation is not a result in people.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    Deciphering Undersegmented Ancient Scripts Using Phonetic Priorabstract

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026 · Texts · Experiment

    The source reads: For Iberian, the method does not show strong evidence supporting Basque as a related language, concurring with the favored position by the current scholarship.

What this does not establish

One model; a negative result is not proof that Iberian has no relative.

Later work looked for

Crossref, 2021 on, "neural decipherment", "undeciphered script" and "Linear A decipherment": no reviewed report of a machine decipherment of a script whose language is unknown; nothing that changes it.

Claim 52 · checked 24 September 2026 · AI-assisted audit

Claim: in line with most current scholarship.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

What this does not establish

The authors’ own characterisation of the field.

Later work looked for

Crossref, 2021 on, "neural decipherment", "undeciphered script" and "Linear A decipherment": no reviewed report of a machine decipherment of a script whose language is unknown; nothing that changes it.

Claim 53 · checked 24 September 2026 · AI-assisted audit

Claim: The bottleneck is material, not cleverness.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s reading of claims 47 to 53, with Sproat’s point about how little can be learned from small corpora.

Claim 54 · checked 24 September 2026 · AI-assisted audit

What would break them

Claim: What would break them

Kind
Framing. Signposting. It asserts nothing a source could check.

Claim: A bilingual inscription, or a long text, or a securely identified related language.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

Follows from the page’s account of what decipherment has needed. Farmer, Sproat and Witzel list a bilingual and a long inscription among the finds that would change their view of the Indus signs.

Claim 56 · checked 24 September 2026 · AI-assisted audit

Claim: All three are matters of excavation and luck rather than of method.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s judgement.

Other readings

A related language is not only a find: Luo and colleagues propose a computational measure of which known language is closest, though on Iberian it found no clear relative (claim 52).

Claim 57 · checked 24 September 2026 · AI-assisted audit

Claim: Linear A is the most promising of the group

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s judgement, resting on the partial reading claims 60 and 62 describe.

  • Context

    SigLA: The Signs of Linear A, a Palaeographical Databasesection 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: we are in a position to be able to at least ‘read’, although with an approximation, and to an extend to interpret inscriptions

Claim 58 · checked 24 September 2026 · AI-assisted audit

Claim: and has not yielded.

Kind
Historical claim. A statement about the past that historical evidence can test.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2021 on, "Linear A" and "decipherment": no accepted decipherment reported; nothing that changes it.

Claim 59 · checked 24 September 2026 · AI-assisted audit

Claim: Its signs are close enough to Linear B’s that the phonetic values of the later script can be read back onto the earlier one,

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

  • Supports

    SigLA: The Signs of Linear A, a Palaeographical Databasesection 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: the phonetic values we know for Linear B signs are retrospectively applied to Linear A homomorphic signs, allowing for an approximate reading of Linear A sign-sequences

What this does not establish

The SigLA authors say that some shared signs are argued to share sounds, and call the reading approximate.

Later work looked for

How the method works; not a finding later work revises.

Claim 60 · checked 24 September 2026 · AI-assisted audit

Claim: which gives sounds without giving a language.

Kind
What scholars argue. A report of what specialists argue, or where they disagree.
Review
Source checked. The cited passage was read, and it says this.

Later work looked for

Crossref, 2021 on, "Linear A" and "Minoan language": nothing that changes it.

Claim 61 · checked 24 September 2026 · AI-assisted audit

Claim: That is what a related script buys on its own: you can roughly pronounce the words and still not know what most of them mean.

Kind
This page’s reading. A judgement the page makes from the evidence. Others read it differently.
Review
A judgement, not a fact. Reviewed for whether it is fairly put and follows from the checked claims. No source can confirm it.

The page’s reading of claims 60 and 61.

  • Context

    SigLA: The Signs of Linear A, a Palaeographical Databasesection 2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: allowing for an approximate reading of Linear A sign-sequences

  • Context

    SigLA: The Signs of Linear A, a Palaeographical Database§ 4.2

    Cited · Link answered, 19 October 2026 · Read in full, 24 September 2026

    The source reads: our knowing the meaning of ku-ro is more of an exception than the rule

    The authors’ own statement that the meaning of Linear A words is mostly unknown, which is what “most of them” rests on.

Corrected

24 September 2026. Previously: “That is what a related script buys on its own: you can pronounce the words and still not know what any of them mean.” The SigLA authors call the reading approximate and say some inscriptions can be interpreted to an extent, so "any of them" overstated it.

Claim 62 · checked 24 September 2026 · AI-assisted audit

What was read