Appendix 2

Cross-tabulation of signs from 28 ancient scripts

Table 1 describes the 28 ancient scripts cross-tabulated in Table 2, irrespectively of the eventual pronunciation or meaning of the signs. The only criterion is the visual similarity of the glyphs. We know that the stance and orientation of the signs of early scripts were not fixed but depended on the direction of writing. Considering this fact and allowing for some minor scribe-dependent variation, I listed similar signs from various scripts in rows. Occasionally, when signs presented various degrees of similarity, I inserted more than one line for the family with the most similar glyphs repeated in separate lines. I gave similar signs the numeric value 1, for simple existence within a script, and identical signs, the value 2. The last column of Table 2, called POP for popularity, gives the sum of the values in a line. Popularity is the number of scripts using a sign. If two or more scripts use the exact same sign, with value 2, the popularity of the sign is thus weighted compared to signs with simply similar versions used in the same number of scripts. For some tests (e.g., chi-square and other non-parametric tests), the entire table takes the values 1 for existence (independently of how similar the signs may look) and 0 for the non-existence of a sign in a script.

Alphabets typically have less than 50 signs. Sets with 50-150 signs are usually classified as syllabic scripts (syllabaries), even those that are admittedly not deciphered. Hieroglyphic systems consist of many more signs, pictograms, ideograms, or logograms with a semantic value but no phonetic value. This kind of syllogism and arithmetic is relatively weak. It makes a western computer keyboard with about 100 keys a syllabic writing tool. Linear B, for example, was given 12 consonants, 5 vowels, and 73-78 syllabic values. All the deciphered signs of Linear B are taken from and are identical to Linear A signs (Salgarella & Castellan, 2021), albeit with different pronunciations. Still, both scripts employ many more signs which remain undeciphered. Some Linear A and B signs are identical to Cretan hieroglyphs, which are thought to have no phonetic or semantic value. Some are even attested in the oldest Archanes script (Decorte, 2018a, 2018b), but the linguistic relevance remains debated. The most ancient signs found in Armenia and the Balkans are thought to have no linguistic relevance in those contexts but acquire one when used in later scripts. So, are Linear A and B truly syllabaries, hieroglyphic systems, or downright alphabets with extensive sign variation?

The similarity between glyphs is challenging to quantify. Some scribes may use distinct strokes, while others would join strokes into fewer curves. How much variation in curvature, angle width, inclination, number, or length of strokes is phonetically and semantically relevant, and how much of it is used for emphasis (e.g., uppercase, lowercase, tonic accent, etc.), punctuation, declination, diacritics, prepositions, particles, other morphemes, or is just random? Fig. 1 shows the progressive abstraction of a Cretan hieroglyphic sign from a pictographic form evoking a funnel-like dispenser with a faucet to a simple K-like letter form. All these, and many more attested intermediate forms, are variants of the same sign (Godart & Olivier, 1996). Suppose we accept the similarity between signs on the left and right. In that case, we cannot dispute the similarity between the signs on the right (Қ-like) and a modern-day K. This K would have a different semantic value from a K derived from a hand-palm sign (see section K). Indeed, in many instances, a K is interpreted as concentrating or dispensing depending on its position in a stem (start or end). Fig. 2 shows various Linear A signs. Each column contains variants of the same sign, and different columns represent different sign classes according to current classification. However, there seems to be as much graphical variation between columns as within.

Figure 1. Variants of Cretan hieroglyphic sign 057 from Godard and Olivier (1996) are arranged with decreasing pictographic complexity from left to right.

Figure 2. Specimens of various Linear A sign classes from Salgarella & Castellan (2021). Columns represent different sign families and rows, specimens that may be confused.

We have the same difficulties when comparing signs between scripts from different geographical regions and times. In Fig. 3, sign 1 is the variant ZA-9 of the Linear A and B sign SigLA AB65 found at Zakros, Crete. Morphologically, this sign correlates very poorly with other members of its family. Instead, it may be considered as a cursive variant of the family of signs SigLA AB09, with mostly angular members, of which upright and rotated examples are shown in positions 2 and 3. When rotated or flipped, sign 1 produces the forms at positions 4-6. Form 4 is identical to sign 7 found in several Byblos inscriptions (Dunand, 1945, 1978) – one example is shown – and in the Phoenician inscription of the Louvre Cippus, which the discoverer of the Phoenician script Jean-Jacques Barthélemy interpreted as a Semitic Mem in 1758 (Barthélémy, 1764). Form 6 is found in several instances in the Ahiram sarcophagus inscription, the oldest specimen of the Phoenician script in its alleged homeland. An extract from this inscription is also shown.

Figure 3. Examples of glyph variation between scripts showing the forms of a sign in Linear A (from Salgarella & Castellan, 2021; 1), the Byblos script (from Dunand, 1945, Fig. 27b; 7), and the Phoenician alphabet (Ahiram sarcophagus inscription; 8). The curved arrows indicate a rotation or flipping.

Table 2 may be used to measure similarity between scripts and testing hypotheses as to their provenance and relationships. We may test whether the Egyptian linear signs descend from the much older Armenian petroglyphs or the Balkan signs. The null hypothesis is that the Egyptian script is independent of the other two. Any resemblance is due to chance. Alternatively, knowing the Balcan or Armenian signs, we can predict the existence of corresponding signs in the Egyptian script. What proportion of our guesses would turn true? The following Chi-square test shows no association of the Egyptian linear signs with the Balkan script (p=0.153; any similarity can be attributed to chance) but a very significant association with the Armenian signs (p=7.9E-5; the null hypothesis is rejected).

		Egyptian linear
	Exists	no		yes
		Count	Row %	Count	Row %
Balkan	no	263	84.6%	48	15.4%
	yes	118	79.2%	31	20.8%
Armenian	no	298	86.9%	45	13.1%
	yes	83	70.9%	34	29.1%

Pearson Chi-Square Tests
		Egyptian linear
Balkan	Chi-square	2.043
	Sig.	.153
Armenian	Chi-square	15.584
	Sig.	7.9E-5*

Results are based on nonempty rows and columns in each innermost subtable. *The Chi-square statistic is significant at the 7.9e-5 level.

We have thus left with the alternative to the null hypothesis that the similarity between the Egyptian and Armenian scripts is not coincidental. There must be a reason other than chance (e.g., learning) to explain that nearly 30% of the Egyptian linear signs also exist among the much older Armenian petroglyphs.

Instead of comparing only two scripts at a time, we can use the ROC curves method to compare a target script to numerous others. We can thus measure the relative contribution of pre-existing scripts to the target script and evaluate the hypothesis of a pure coincidence for each presumed contributor. For example, did any Armenian, Balkan, Egyptian, Cretan hieroglyphic, Cypro-Minoan, or Proto-Sinaitic scripts significantly influence Linear A, and how much? This test plots the true positive rate (Sensitivity) over the false positive rate (1-Specificity) for each predictor. The further away the curve departs from the diagonal towards the upper-left corner, the better the predictor is. 

Fig. 4 shows the results for the above question. Linear A is mainly influenced by the Balkan script (p=5.6e-3), but it also significantly resembles Armenian and Proto-Sinaitic (p<0.05). Surprisingly, it seems independent from Cretan hieroglyphs. Any similarities between these two Cretan scripts are due to chance or random archeological artifacts. 

Figure 4. ROC curves comparing various older or co-existing scripts as progenitors of Linear A.

For securely dated scripts, we can tell which is the progenitor and which is the progeny. Linear A cannot have influenced the Balkan script because the latter is much older. Only the opposite proposition makes sense. With the Sinaitic script being approximately contemporary to Linear A, the resulting significant similarity may be interpreted in either direction. The above model measures how much the Sinaitic script, among others, has influenced Linear A. We may invert the model and ask how much Linear A has influenced the Sinaitic script. The results in Fig. 5 suggest that the Linear A > Sinaitic hypothesis (p=0.018) fits the data better than the Sinaitic > Linear A model (p=0.041). More precisely, the two scripts appear to have different origins, the Sinaitic being an Armenian-born script (p=1.2e-6) while Linear A is more of a Balkan script (p=5.6e-3).

Figure 5. ROC curves comparing Linear A and various older or co-existing scripts as progenitors of the Proto-Sinaitic scripts.

The ROC-curves method puts numbers to an obvious syllogism. Suppose you find 22 different objects in my house and the same items in a large supermarket among hundreds of other objects. the assumption that the supermarket purchases its products from my house is awkward; rather, I buy my provisions from the supermarket. A large set of distinct characters, like the Linear A script, cannot derive from a small subset, such as the Proto-Sinaitic script. The small subset more likely derives from a larger set.

References

Barthélémy, J.-J. (1764). Réflexions sur quelques monumens phéniciens et sur les alphabets qui en résultent. Histoire de l’Académie Royale Des Inscriptions et Belles-Lettres, 30, 405. 

Decorte, R. P.-J. E. (2018a). The First ‘European’ Writing: Redefining the Archanes Script. Oxford Journal of Archaeology, 37(4), 341–372. 

Decorte, R. P.-J. E. (2018b). The Origins of Bronze Age Aegean Writing: Linear A, Cretan Hieroglyphic and a New Proposed Pathway of Script Formation. In S. Ferrara & M. Valério (Eds.), Paths into Script Formation in the Ancient Mediterranean (pp. 13–49). Edizioni Quasar. 

Dunand, M. (1945). Byblia Grammata. Documents et recherches sur le développement de l’écriture en Phénicie, Beirut (Vol. 2). Libanese Ministry of National Education and Fine Arts. 

Dunand, M. (1978). Nouvelles inscriptions pseudo-hiéroglyphiques découvertes à Byblos. Bulletin Du Musée de Beyrouth, 30, 51–59.

Godart, L., & Olivier, J.-P. (1996). Corpus Hieroglyphicarum Inscriptionum Cretae. In Études crétoises (Vol. 31). École française d’Athènes. 

Salgarella, E., & Castellan, S. (2021). SigLA: The Signs of Linear~A. A~Palæographical Database. Grapholinguistics in the 21st Century, Part II, 5, 945–962. 

Table 1: Description of the scripts cross-tabulated in Table 2.

Call	Full name of the script	Proposed language	Attestation region	Earliest
ARM	Armenian petroglyphs	not a language	Armenia	7000 BC
BAL	Balkan signs	not a language	Balkan Peninsula	6000 BC
EGY	Egyptian (linear signs only)	Egyptian	Egypt	3000 BC
CHI	Cretan hieroglyphs	unknown language	Crete	2100 BC
LIA	Linear A	unknown language	Crete, Aegean, Anatolia	1800 BC
LIB	Linear B	Greek	Crete and mainland Greece	1450 BC
CYM	Cypro-Minoan and Cypriot	Greek	Cyprus	1550 BC
SIN	Proto-Sinaitic	West Semitic	Sinai Peninsula and Egypt	1900 BC
CAN	Canaanite	West Semitic	Canaan	1800 BC
BYB	Byblos	Unknown language	Byblos	1800 BC
PHO	Phoenician	West Semitic	Mediterranean (mainly western)	1050 BC
GRE	Archaic Greek (pooled)	Greek	Aegean and mainland Greece	750 BC
LYD	Lydian	Luwian-related (IE)	Western Anatolia	700 BC
LYC	Lycian	Luwian-related (IE)	Western Anatolia	500 BC
PHR	Phrygian	Greek-related (IE)	Central Anatolia	800 BC
CAR	Carian	Greek-related	South-West Anatolia	700 BC
PAM	Pamphylian	Greek	South Anatolia	500 BC
MON	Old Turkic	Turkic	Mongolia	700 AD
SAR	Old South Arabian	South Semitic	South Arabian Peninsula	800 BC
ETR	Etruscan	Old Italic	Central Italy	700 BC
LEP	Lepontic	Old Italic/Celtic	Northern Italy	550 BC
NUC	Nucerian	Old Italic	Southern Italy	600 BC
OSC	Oscan	Old Italic	Southern Italy	500 BC
CAM	Camunic	Old Italic/Germanic	Northern Italy	500 BC
RHA	Rhaetic	Old Italic/Etruscan	Northern Italy	500 BC
VEN	Venetic	Old Italic/Germanic	Northern Italy	600 BC
RUN	Runic	Germanic	Central Europe	150 AD
GLA	Glagolitic	Slavic	Central and East Europe	800 AD

Table 2: Cross-tabulation of 28 ancient scripts (click to magnify).