data/linear_a/METHODOLOGY.md

# Methodology: Linear A Libation Formula Signs Dataset

## 1. Collating the Inscriptions

### 1.1 Source PDFs

Two academic books by Ester Salgarella were the primary textual sources:

**PDF 1 — Salgarella 2024, "Writing in Bronze Age Crete" (Cambridge Elements, 84 pp.)**
- Tables 5-6 (pp. 51-52): The libation formula structure, parsed into 6 positional slots (verb, place-name, dedicant, object, subordinate verb, prepositional phrase). IO Za 2 is the principal exemplar with full transcription: `a-ta-i-*301-wa-ja ja-di-ki-tu [dedicant] ja-sa-sa-ra-me u-na-ka-na-si i-pi-na-ma si-ru-te`.
- Section 7.2 (pp. 45-46): A 301 discussed as the central verb root, with variant forms listed (a-ta-i-*301-wa-ja, a-ta-i-*301-wa-e, ta-na-i-*301-ti, ta-na-i-*301-u-ti-nu).
- Section 4 (pp. 20-21): The AB/A naming convention explained. AB = shared between Linear A and Linear B. A = Linear A only.
- Section 6.2 (pp. 38-41): Stone vessels (Za category) described as cultic/ritual. 9 Za inscriptions with GORILA references cited.
- Section 8.1 (p. 50): Phonetics policy — reading LA via LB values is "approximate." The complete sign list is deferred to the online appendix at www.cambridge.org/salgarella (inaccessible, returns 403).
- p. 49: AB 07/DI and AB 28/I explicitly named with their readings.
- p. 57: AB 30/NI confirmed as logogram for "fig" (acrophonic: Minoan *nikyleon).

**PDF 2 — Salgarella 2020, "Aegean Linear Script(s)" (Cambridge Classical Studies)**
- Table 2 (p. 35): The core CV syllabary grid — the complete mapping of consonant-vowel combinations to AB numbers. This was the primary source for all 32 deciphered sign readings.
- Table 3 (p. 36): Additional syllabary signs (a2, a3, au, dwe, dwo, nwa, pte?, pu2, ra2, ra3, ro2, ta2?, twe?, two).
- Table 4 (p. 37): **Undeciphered syllabograms**: *18, *19, *22, *34/35, *47, *49, **\*56**, *63, *64, *65, *79, *82, *83, *86, *89. This is where AB 56 is explicitly classified as undeciphered.
- Table 5 (pp. 65-68): Catalog of all non-administrative inscriptions by site, including Za entries. This was the basis for the initial inscription inventory.
- pp. 33-34: The homomorphy-homophony principle and evidence hierarchy for phonetic values (Cypriot cross-script confirmation > shared toponyms > acrophonic logograms).
- p. 234, p. 355: SY Za 4 cited as the sole Linear A attestation of sign nwa (AB 48).
- Index of Signs (pp. 412-416): Complete list of every AB and A sign cited in the book, with page references.

### 1.2 SigLA Database (sigla.phis.me)

SigLA (Signs of Linear A) is an online palaeographic database by Salgarella & Castellan, licensed CC BY-NC-SA 4.0. The SSL certificate was expired during scraping, so all access used `curl -sk` to bypass verification.

**Discovery method:**
- The browse page (`/browse.html`, 772 documents) was searched for all Za entries. Only 2 found: PS Za 2 and SY Za 4.
- The kind pages were probed: `/kind/Libation%20table/` returned PS Za 2 and SY Za 4. All other kind names (Stone vessel, Ladle, Bowl, etc.) returned 404.
- Individual document pages were probed for all 60 Za inscriptions in our database. Only PS Za 2 and SY Za 4 returned 200; all others returned 404.

**Sign data extraction from SigLA:**
- PS Za 2: 16 signs extracted from `/document/PS%20Za%202/` and individual sign pages (`/document/PS%20Za%202/index-1.html` through `index-16.html`).
- SY Za 4: 13 signs extracted from `/document/SY%20Za%204/` and individual pages.
- Each sign page contains: AB number in a `sure-reading` or `info-title` span, conventional reading, sign type (syllabogram), reading certainty, and an individual sign image PNG URL.
- Word boundaries were extracted from `/document/PS%20Za%202/index-word.html` and `/document/SY%20Za%204/index-word.html`, which use `class="word"` dividers.

### 1.3 lineara.xyz (Digital GORILA Corpus)

The JavaScript data file `LinearAInscriptions.js` was the definitive ground-truth source for inscription existence and Unicode sign data. Accessed via `curl -sk "https://lineara.xyz/LinearAInscriptions.js"`.

**Discovery method:**
- The file contains a JavaScript `Map` with every inscription as a key-value pair.
- All Za entries were extracted with: `grep -oP '^\["[A-Z]+Za[^"]*"'`
- The sitemap.xml was cross-verified to ensure no entries were missed.
- Result: 59 unique Za inscription keys (including angle-bracket uncertain attributions like `IO Za <1>` and `AP Za <3>`).

**What this provided:**
- `parsedInscription`: Unicode Linear A characters for each inscription
- `transliteratedWords`: Syllable-by-syllable transliteration (e.g., `"A-TA-I-*301-WA-JA"`)
- `words`: Unicode character sequences per word (e.g., `"𐘇𐘳𐘚𐙕𐘮𐘱"`)
- `site`, `support`, `context` metadata
- Image URLs for facsimiles and inscription photographs

### 1.4 Audit Against lineara.xyz

The initial database (v1.0.0, 45 entries) was built from the two PDFs + SigLA. A deep audit against lineara.xyz revealed:
- **v1.1.0**: Added 13 entries from GORILA digital corpus (IO Za 3/4/5/14, SY Za 5, KY Za 2, KAN Za 1, LACH Za 1, KN Za 13/17/18/19, KO Za 4). Removed none.
- **v1.2.0**: Further verification against `LinearAInscriptions.js` map keys revealed 2 hallucinated entries (KN Za 13, KO Za 4 — an earlier agent had fabricated these) and 4 genuinely missing entries (IO Za 13, IO Za 15, NE Za 1, ZO Za 1). Final: 60 entries matching 59/59 lineara.xyz entries plus IO Za 10 from Salgarella 2020.

---

## 2. Converting Signs to Unicode Characters

### 2.1 The Unicode Linear A Block

Linear A signs are encoded in Unicode block U+10600–U+1077F (Linear A, added in Unicode 7.0, 2014). Each sign has a codepoint named `LINEAR A SIGN AB0XX` or `LINEAR A SIGN AXXX`.

### 2.2 Mapping Method

The mapping from conventional reading to Unicode codepoint was established by **aligning transliterated words with their Unicode character sequences** from lineara.xyz:

**Step 1: Extract paired data.** For IO Za 2, lineara.xyz provides:
```
transliteratedWords: ["A-TA-I-*301-WA-JA", ...]
words:              ["𐘇𐘳𐘚𐙕𐘮𐘱", ...]
```

**Step 2: Split and align.** Each transliterated word is split on hyphens (handling `*NNN` as single tokens). Each Unicode word is split into individual characters. The resulting parallel arrays give a 1:1 syllable-to-codepoint mapping:
```
A    → 𐘇 (U+10607)
TA   → 𐘳 (U+10633)
I    → 𐘚 (U+1061A)
*301 → 𐙕 (U+10655)
WA   → 𐘮 (U+1062E)
JA   → 𐘱 (U+10631)
```

**Step 3: Verify across multiple words.** The same mapping was verified across all 6 words of IO Za 2, plus PS Za 2 and SY Za 4, ensuring consistency. The character `𐝫` (U+10777, LINEAR A SIGN UNDETERMINED) was filtered out as a damage/lacuna marker.

**Step 4: Cross-check with SigLA.** The AB numbers from SigLA's sign data (e.g., "AB 59" for position 1 of PS Za 2) were verified against the Unicode codepoint names (e.g., U+10633 = LINEAR A SIGN AB059). All 29 signs present in both SigLA and lineara.xyz matched perfectly.

### 2.3 Signs Verified

| Reading | AB Number | Unicode | Verification Source |
|---------|-----------|---------|---------------------|
| a | AB 08 | U+10607 𐘇 | IO Za 2 word 1, SY Za 4 word 1 |
| ta | AB 59 | U+10633 𐘳 | IO Za 2 word 1, PS Za 2 word 1, SY Za 4 word 1 |
| i | AB 28 | U+1061A 𐘚 | IO Za 2 word 1, PS Za 2 word 1, NE Za 1 |
| *301 | A 301 | U+10655 𐙕 | IO Za 2 word 1, PS Za 2 word 1, SY Za 4 word 1 |
| wa | AB 54 | U+1062E 𐘮 | IO Za 2 word 1, SY Za 4 word 1 |
| ja | AB 57 | U+10631 𐘱 | IO Za 2 words 1/2/4, PS Za 2 words 2/3, SY Za 4 words 1/2 |
| di | AB 07 | U+10606 𐘆 | IO Za 2 word 2 |
| ki | AB 67 | U+10638 𐘸 | IO Za 2 word 2 |
| tu | AB 69 | U+10639 𐘹 | IO Za 2 word 2 |
| sa | AB 31 | U+1061E 𐘞 | IO Za 2 word 4, PS Za 2 word 4 |
| ra | AB 60 | U+10634 𐘴 | IO Za 2 word 4, PS Za 2 word 4 |
| me | AB 13 | U+1060B 𐘋 | IO Za 2 word 4, PS Za 2 word 4 |
| u | AB 10 | U+10609 𐘉 | IO Za 2 word 5 |
| na | AB 06 | U+10605 𐘅 | IO Za 2 word 5, PS Za 2 word 1 |
| ka | AB 77 | U+1063E 𐘾 | IO Za 2 word 5 |
| si | AB 41 | U+10624 𐘤 | IO Za 2 words 5/6 |
| pi | AB 39 | U+10622 𐘢 | IO Za 2 word 5 |
| ma | AB 80 | U+10641 𐙁 | IO Za 2 word 5 |
| ru | AB 26 | U+10618 𐘘 | IO Za 2 word 6 |
| te | AB 04 | U+10603 𐘃 | IO Za 2 word 6 |
| ti | AB 37 | U+10620 𐘠 | PS Za 2 word 1, SigLA position 5 |
| re | AB 27 | U+10619 𐘙 | PS Za 2 word 5, SigLA position 14 |
| ke | AB 44 | U+10625 𐘥 | PS Za 2 word 5, SigLA position 16 |
| nwa | AB 48 | U+10629 𐘩 | SY Za 4 word 2 |
| za | AB 17 | U+1060D 𐘍 | SY Za 4 word 2 |
| *56 | AB 56 | U+10630 𐘰 | SY Za 4 word 3 (lineara.xyz reads as PA₃) |
| ni | AB 30 | U+1061D 𐘝 | SY Za 4 word 3 |
| wi | AB 40 | U+10623 𐘣 | SY Za 4 word 3 |
| da | AB 01 | U+10600 𐘀 | NE Za 1 (I-DA = 𐘚𐘀) |
| e | AB 38 | U+10621 𐘡 | Standard grid (variant suffix -wa-e) |
| nu | AB 55 | U+1062F 𐘯 | IO Za 2 (variant verb -u-ti-nu) |
| ri | AB 53 | U+1062D 𐘭 | Standard grid (toponym tu-ri-sa) |
| se | AB 09 | U+10608 𐘈 | Standard grid (toponym se-to-i-ja) |
| to | AB 05 | U+10604 𐘄 | Standard grid (toponym se-to-i-ja) |

---

## 3. Assigning Phonetic Values via Linear B Homophony

### 3.1 The Homomorphy-Homophony Principle

The standard method (Salgarella 2020 pp. 33-34; 2024 p. 50) works as follows:

1. **Identify homomorphs**: If a Linear A sign is graphically identical or very similar to a Linear B sign, they are "homomorphs" and receive the same AB number.
2. **Transfer the LB reading**: Since Linear B is deciphered (Ventris 1952), its known phonetic values are applied to the matching Linear A signs as "approximate" readings.
3. **Qualify as "approximate"**: Salgarella repeatedly emphasizes these are approximate — we are explaining an earlier phonological system using a later one.

Of the ~87 distinct syllabograms in Linear B, approximately 59-61 form the "core syllabary" with securely deciphered values. About 14 remain undeciphered (*18, *22, *34/35, *47, *49, **\*56**, *63, *64, *65, *79, *82, *83, *86, *89). The rest are "additional syllabary" signs with specialized values (nwa, pu2, ta2, etc.).

### 3.2 Three Tiers of Phonetic Confidence

We classified each sign's phonetic status using Salgarella's evidence hierarchy (2020 pp. 33-34):

**Tier 1 — "Secure" (11 signs):** Confirmed by **cross-script comparison with the Cypriot Syllabary** (an independently deciphered descendant of the Aegean script tradition). Salgarella 2020 p. 34 lists these signs as having values confirmed via the Cypriot connection: **a, i, da, na, pa, po, ro, sa, se, ti, to**. Of these, the following appear in our libation formula corpus: **a** (AB 08), **i** (AB 28), **da** (AB 01), **na** (AB 06), **sa** (AB 31), **se** (AB 09), **ti** (AB 37), **to** (AB 05).

Three additional signs were upgraded to "secure" based on supplementary evidence:
- **di** (AB 07): Explicitly named as "AB 07/DI" in Salgarella 2024 p. 49. Also confirmed via the shared toponym DI-KI-TE (Mount Dikte), which appears in both Linear A and Linear B.
- **ni** (AB 30): Explicitly confirmed as the logogram for "fig" via acrophonic principle (Minoan *nikyleon), per Salgarella 2024 p. 57.
- **ta** (AB 59): Confirmed via shared personal names and toponyms across LA/LB.

**Tier 2 — "Approximate" (21 signs):** AB signs shared between LA and LB whose readings are applied from LB based on the homomorphy-homophony principle, but lack independent confirmation from Cypriot or other external sources. These include: ja, wa, ki, tu, ra, me, u, ka, si, pi, ma, ru, te, re, ke, nwa, za, wi, e, nu, ri. These readings are used universally in scholarly transcription and are considered reliable enough for practical use.

**Tier 3 — "Undeciphered" (2 signs):** Signs with no established phonetic value. See Section 5 below.

---

## 4. SigLA Image Pipeline

For the two inscriptions with full SigLA data, individual sign images are available as PNG files:

**URL pattern**: `https://sigla.phis.me/document/{INSCRIPTION}/{INSCRIPTION}_{POSITION}.png`

Example: Sign 1 of PS Za 2 → `https://sigla.phis.me/document/PS%20Za%202/PS%20Za%202_1.png`

These URLs are stored in the `sigla_image_urls` field of each sign entry. 18 of the 34 unique signs have at least one SigLA image URL (the ones attested in PS Za 2 or SY Za 4). The remaining 16 signs (those only attested in IO Za 2 or variant/toponym forms) do not have SigLA images because SigLA only contains PS Za 2 and SY Za 4.

For lineara.xyz, facsimile images are available at `https://lineara.xyz/images/{ID}-Facsimile.jpg` and inscription drawings at `https://lineara.xyz/images/{ID}-Inscription.jpg` (e.g., `IOZa2-Inscription.jpg`). These show whole inscriptions, not individual signs.

---

## 5. The Two Undeciphered Signs

### 5.1 Sign A 301 (*301) — U+10655 𐙕

**What it is:** An anthropomorphic pictogram depicting a human figure bending toward and grasping a vertical pole or staff. Informally called the "slave sign" or "acrobat sign." Descended from Cretan Hieroglyphic CH 046 (traditionally described as an "adze with handle").

**Why it is undeciphered:**

**(a) It is Linear A-only.** Sign A 301 belongs to the 300-series — signs that exist ONLY in Linear A with no Linear B counterpart. The standard decipherment method (transfer LB reading → LA) is therefore impossible. When Mycenaean Greeks adapted Linear A into Linear B (~1450 BCE), they did not carry this sign over, probably because it represented a Minoan phoneme absent from Greek.

**(b) It is confirmed as a syllabogram** (not merely an ideogram). Evidence:
- It appears **within polysyllabic words** flanked by other syllabograms (e.g., A-TA-I-**\*301**-WA-JA), carrying prefixes and suffixes that indicate verbal morphology (Davis 2013).
- It participates in **ligatures** (composite signs): A 503 (DA+301), A 519/A 520 (I+301), A 553 (301+ME). Logograms do not typically form ligatures with syllabograms.
- However, it ALSO functions as a **logogram** when written standalone on seals and nodules (235 out of 287 attestations are standalone on Haghia Triada nodules). This dual function is itself an obstacle — it is unclear in some contexts whether the sign is contributing a syllable or a semantic unit.

**(c) Proposed values remain contested.** At least six readings have been proposed:

| Proposal | Proponent | Basis | Problem |
|----------|-----------|-------|---------|
| **jo** | Raison & Pope 1977 | Visual similarity to LB *36 (jo) rotated 90° | Evidence suggests *301 ends in -U, not -O |
| **do** | Solca | Formal comparison | DO is already securely assigned to LB *14 / LA sign 101 |
| **we** | Minoablog | Late Minoan variants converge on LB *75 (we) form | Would give a-ta-i-we-wa-ja — unusual semivowel cluster |
| **ri/li** | Samson 2021 | Computational alignment; visual similarity to LB ri | Assumes Greek substrate (minority position) |
| **zu** | Minoablog (alt.) | Etruscan-Lemnian *AIS ("god") connection | Etruscan-Minoan link is speculative |
| **DYAU** | Indo-European hypothesis | Parallel with Luwian *tatis Tiwaz* | Treats *301 as logogram, not syllabogram |

None has gained scholarly consensus. The sign remains transcribed as `*301` with an asterisk.

**(d) Attestation statistics** (from lineara.xyz `LinearAInscriptions.js`):
- 287 total attestations across 16+ sites
- 235 are standalone on Haghia Triada nodules (administrative/sealing)
- 19 on stone vessels (libation formula context)
- 15 on tablets
- 11 on roundels
- The word `A-TA-I-*301-WA-JA` appears 11 times across sanctuary sites (Iouktas, Syme, Kophinas, Palaikastro, Troullos, Zakros)

### 5.2 Sign AB 56 (*56 / PA₃) — U+10630 𐘰

**What it is:** An abstract linear sign shared between Linear A and Linear B (hence the AB prefix). Unlike A 301, this sign DOES exist in Linear B — but it is undeciphered in Linear B too.

**Why it is undeciphered:**

**(a) Undeciphered in Linear B itself.** AB 56 is one of approximately 14 Linear B syllabograms whose phonetic values have never been formally established. The Comité International Permanent des Études Mycéniennes (CIPEM) has not ratified any reading. It is always transcribed as `*56` in formal scholarship.

**(b) Too rare for definitive analysis.** The standard LB decipherment method relies on spelling alternations (the same word spelled different ways, allowing triangulation). Sign *56 appears in only ~33 inscriptions in Linear A and is similarly rare in Linear B, providing insufficient data for conclusive alternation analysis (Judson 2020, *The Undeciphered Signs of Linear B*, Cambridge UP).

**(c) It is confirmed as a syllabogram** (not an ideogram). Evidence:
- It appears within multi-sign words in both LA and LB: `*56-ra-ku-ja` (LB textile term), `KU-PA₃-NU` (LA commodity term, 8 attestations), `*56-ni-wi` (SY Za 4 libation formula).
- Unicode classifies it as a "SYMBOL" rather than "SYLLABLE" only because its value is unknown, not because it is ideographic.

**(d) The leading candidate reading is PA₃ (/pha/) but it is unconfirmed.** The evidence:

| Evidence | For pa3 | Against pa3 |
|----------|---------|-------------|
| LB textile term `*56-ra-ku-ja` alternates with `pa-ra-ku-ja` | Strong: same word, different spelling → *56 ≈ pa | Could be different words or a scribal error |
| LB personal name `*56-ru-we` vs. `ko-ru-we` | Weak: if same person → *56 ≈ ko, NOT pa | May be different individuals with similar names |
| Structural prediction: pu2 = /phu/ exists, so pa3 = /pha/ should exist | Supports pa3 as filling a predicted gap | Prediction ≠ proof |

Judson (2017, 2020) considers pa3 (/pha/) "most likely" but explicitly states there is insufficient evidence for formal acceptance. Melena (1987, 2000) argued that the `pa-ra-ku-ja` / `*56-ra-ku-ja` alternation had "failed" as proof — the alternations were not robust enough.

**(e) lineara.xyz uses PA₃ as a working convention.** The digital GORILA corpus transcribes *56 as `PA₃` in all contexts, which is why the word on SY Za 4 appears as `PA₃-NI-WI` rather than `*56-NI-WI`. This is a practical convention, not a scholarly consensus.

**(f) It likely represents a Minoan phoneme.** Melena (1987, 2000) argued that *56 and similar undeciphered signs represent **pre-Greek Minoan consonants** — sounds that existed in the Minoan language but not in Mycenaean Greek. When Greeks used Linear B, these signs appeared only sporadically in pre-Greek names and loanwords, making them intrinsically harder to decipher.

**(g) Attestation statistics** (from lineara.xyz):
- 33 inscriptions, 39 word tokens, 28 distinct word forms
- Most common word: `KU-PA₃-NU` (8 occurrences, Haghia Triada/Phaistos tablets)
- Libation formula: appears once, in `PA₃-NI-WI` on SY Za 4
- Distributed across 8 sites including Miletos (Anatolia) — one of few LA signs attested outside Crete

---

## 6. Summary

| Category | Count | Signs |
|----------|-------|-------|
| Total unique signs in libation formula | 34 | |
| Secure phonetic value (Cypriot + other confirmation) | 11 | a, i, da, di, na, ni, sa, se, ta, ti, to |
| Approximate phonetic value (LB homophony) | 21 | ja, wa, ki, tu, ra, me, u, ka, si, pi, ma, ru, te, re, ke, nwa, za, wi, e, nu, ri |
| Undeciphered | 2 | *301 (LA-only, no LB match), *56 (undeciphered in LB too) |

Both undeciphered signs are **confirmed syllabograms** (not ideograms) based on their syntactic behaviour within polysyllabic words. They resist decipherment for fundamentally different reasons:

- **\*301** has no Linear B counterpart at all — the standard decipherment pathway is completely blocked.
- **\*56** exists in Linear B but is too rare there to be deciphered — the standard pathway exists but lacks sufficient data.

These are the two signs that the PhoneticPriorModel would need to assign phonetic values to. *301 is the more impactful target: it appears ~287 times in the corpus and forms the core verb root of the libation formula. *56 appears ~33 times and occurs in a single libation formula word.

---

## References

- Salgarella, E. (2024). *Writing in Bronze Age Crete*. Cambridge Elements.
- Salgarella, E. (2020). *Aegean Linear Script(s)*. Cambridge Classical Studies. Tables 2-4 (pp. 35-37), Index of Signs (pp. 412-416).
- Davis, B. (2013). Syntax in Linear A: The word-order of the 'libation formula'. *Kadmos* 52(1), pp. 35-52.
- Davis, B. (2014). *Minoan Stone Vessels with Linear A Inscriptions*. Aegaeum 36.
- Judson, A. (2020). *The Undeciphered Signs of Linear B*. Cambridge UP.
- Judson, A. (2017). The mystery of the Mycenaean 'labyrinth': the value of Linear B pu2. *SMEA*.
- Melena, J. (1987). On Untransliterated Syllabograms *56 and *22. *Tractata Mycenaea*, pp. 203-231.
- Melena, J. (2000). On the Structure of the Mycenaean Linear B Syllabary: I. The Untransliterated Syllabograms.
- Muhly, P. & Olivier, J.-P. (2008). Linear A inscriptions from the Syme Sanctuary. *Archaiologike Ephemeris* 147, pp. 197-223.
- SigLA database: https://sigla.phis.me/ (CC BY-NC-SA 4.0)
- lineara.xyz — Digital GORILA corpus by George Douros
- GORILA = Godart & Olivier (1976-1985). *Recueil des inscriptions en lineaire A*. Vols. I-V.