Volume 94 1985 > Volume 94, No. 4 > On a fragment of the 'Tahua' tablet, by J. B. M. Guy, p 367-388
ON A FRAGMENT OF THE “TAHUA” TABLET
Part of the first line of the tablet “Tahua” (or the “Oar”) in the Museum of the Congrégation des Sacrés Coeurs de Picpus in Braine-le-Comte (Belgium) will be shown here to be a variant of a text found on three other tablets. Minor improvements on Barthel's alphanumerical notation of the glyphs are proposed, and guidelines for the analysis of these parallel texts elaborated and justified. The conclusions drawn from the subsequent analysis of the texts along these guidelines throw some light on the nature and the history of the Easter Island writing.
The same texts often occur, with minor variations, on different tablets of Easter Island. This fact has been highlighted in Barthel (1958a:57-70), in which the alphanumerical transcriptions of tablets G and K, and of tablets H, P and Q (in Barthel's system of reference) are paralleled.
Tablets H, P and Q have no Pascuan names. Tablet H, called the “Grande Tablette”, was collected in January 1870 and presented to the National Museum of Natural History of Santiago (Stephen-Chauvet 1935:74). Tablets P and Q, variously referred to as “tablette A” and “tablette B”, “MAE II” and “MAE I”, or “Grosse Leningradtafel” and “Kleine Leningradtafel” (Barthel 1958a:28-9), were collected in June-July 1871 (Stephen-Chauvet 1935:74). Tablet Q is reported by Barthel to be incomplete.
A fragment of the text which occurs on the fifth and sixth lines of the recto of H and Q and on the fifth line of the recto of P is also found, with considerable stylistic variations and two groups of glyphs transposed, on the first line of side A of the tablet “Tahua”, also called “la Rame”, or the “Oar” (Tablet A in Barthel's nomenclature).
Tablet A is engraved on a European oar of ash wood and is considered to date from the late 18th century or the first half of the 19th century, whereas the other tablets are inscribed on local woods (Metraux 1940:393).
Figures 1 to 4 show the text in its four variants with some of its context. Each glyph is referenced by three sets of figures: (1) its position in the text, the first glyph of the line where the text occurs being assigned the number 1 (ligatures are counted as one glyph); (2) its transcription as per Barthel - 368 1958a:44, 66-7; and (3) its transcription using Barthel's system, slightly altered.- 369 - 370 - 371 - 372
The reasons for tampering with Barthel's transcription are as follows:
Referencing Glyphs and Textual Fragments
Textual fragments will be referenced, as convenience dictates, in any of two ways:
METHODS AND AIMS OF THE ANALYSIS
A purely structural analysis of the text common to tablets A, H, P and Q is sought. No attempt at translation is to be made, any hypotheses about the possible functions and meanings of the glyphs are to be carefully avoided, and no use is to be made in this analysis of the data obtained by various authors from Pascuan informants (e.g., Metoro, Teao, Te Haha, Urevaeiko) or to studies of Pascuan and Polynesian culture and traditions (Metraux 1940, Barthel 1960, etc.), for it is felt that, given the justifiably low level of acceptance with which any hypotheses about the Easter Island tablets are likely to be met, resorting to such external evidence could only weaken whatever conclusions might be drawn from the analysis of these four parallel texts.
Principles of Segmentation
By segmentation is meant the division of the text into significant parts. The previous sentence may be used for an example of precisely what is involved in the process. A possible partition of it is: By/ segmentation/ is/ meant the division/ of the text/ into/ significant parts/. Another is: By seg/ mentation is me/ ant the div/ ision of the t/ ext into sig/ nificant parts. The former partition is clearly vastly better than the latter, even though the text is divided sometimes into single words, sometimes into groups of words which do not necessarily form a complete phrase. But is it so clearly better? Not in the least. The choice of the first partition as superior is backed by the knowledge of the language of the text. Given a text of similar length in an unknown language and script, both partitions would appear equally arbitrary: whether spaces mark word boundaries or are phonetic symbols would not even be known. The automatic segmentation of continuous texts has been attempted (Suhotin 1960), but the procedures are computationally extremely expensive and the results disappointing (Guy 1977: computer tape). Until more reliable algorithms have been discovered, the Easter Island data should not be submitted to these computational techniques.
The method used here is inspired from those of textual criticism.
Imagine the sentence: By segmentation is meant the division of the text into significant parts. to be part of a manuscript of which many versions are extant. It may be missing altogether in some versions. It may be misplaced. Parts of it may be altered: the word segmentation replaced by partition, significant by meaningful, parts by components, is meant by we mean. In all cases these substitutions, omissions, and transpositions affect mean- - 374 ingful segments of the text. Only a copyist with no understanding of the manuscript could omit or transpose a meaningless segment such as “tation is me”, or replace it with some other sequence of letters, meaningful or not. But would such a copyist want to edit the text at all? Editing a text supposes some understanding of it, even if imperfect, and copyists given texts they do not understand can be expected to reproduce them as faithfully as their skill allows rather than alter them.
Although alterations do affect meaningful segments of a text, care is required in identifying what constitutes a significant segment from the evidence of the variants encountered, as the mere listing of the variants will not necessarily restore the exact changes effected by the copyist. The two versions of the sentence in the example read, after carrying out the changes mentioned above: By segmentation is meant the division of the text into significant parts; and: By partition we mean the division of the text into meaningful components. Their inspection for variants gives the following correspondences: segmenta: parti, we: is, (zero): t, significant par: meaningful componen. Note that the fragments involved in these correspondences are all contained in meaningful segments. This is not fortuitous.
Call “string” a sequence of symbols, “substring” a sequence of symbols occurring in a string, e.g., seq, quence of s, and symbols are substrings of a sequence of symbols. Note that the definition of substring implies that any string is a substring of itself. Then, since textual alterations affect meaningful segments of text, variants are necessarily substrings of meaningful segments. Call them “cores”. The segmentation of a text according to the variants encountered will therefore result in strings of the type LCR LCR LCR in which C is a core identified from correspondences between variants, and L and R are the left and right remainders of significant segments, and may be null strings.
Consider a manuscript M containing a meaningful segment A in an environment X . . . Y (i.e., XAY). Manuscript M' is a version of M′ in which the copyist has replaced A with segment B (i.e., XBY). Since alterations affect meaningful segments, B is also a meaningful segment. Comparing the two manuscripts should in theory yield the variant pair A/B, both members of which are meaningful segments.
Imagine strings A and B to consist respectively of iaj and vaw. The fragments compared read therefore XiajY and XvawY, causing the identification of two variant pairs i/v and j/w which do not correspond to any meaningful segments. This misidentification was caused by the fortuitous occurrence of substring a within segments A and B.
Consider now the case in which A and B consist of aiz and ajz. The fragments of M and M′ now read XaizY and XajzY and yield the variant pair i/j which, again, does not correspond to the original meaningful - 375 segments. The misidentification is again caused by the fortuitous resemblance of the leading and trailing substrings of A and B.
Imagine a text written in a system making use of N different symbols in roughly equal proportions. There is one chance in N that any string of one symbol in a given text may fortuitously be found in the corresponding position of another text, and generally one chance in about N to the power L that any string L symbols long of a text may be found in the corresponding position of another text by chance alone.
Therefore, the greater the number of different symbols in a writing system, the less chance of variant pairs being incomplete substrings of meaningful segments and, consequently, the shorter the average length of the remainders and the longer the lengths of the cores.
Moreover, since the probability of the occurrence of chance correspondences is roughly inversely proportional to the number of different symbols to the power of the length of the variant pairs involved, the longer the variant pairs, the greater their chance of corresponding to whole meaningful segments, and the shorter the average length of their remainders.
It follows from these observations that a partition of a text which results in short segments is more likely to be erroneous than one resulting in long segments, and that the identification of pair variants should be carried out by considering only long textual fragments. The smaller the number of different symbols in the writing system, the longer these fragments should be.
Consider again the partly erroneous segmentation of the two versions of the same sentence above. It leads to the following partitions: /By /parti/ tion /we/ mean// the division of the text into /meaningful par/ts./ and: /By /segmenta/tion /is mean/t/ the division of the text into / significant componen/ts./ Only one of these segments (ts.) does not correspond to a morpheme, and both partitions are characterised by the presence of very short segments (By, t, ts.). Disallowing segments less than four characters long would have forced a better partition: /By segmenta/tion /is meant /the division of the text into / meaningful parts./ and: /By parti/ tion /we mean /the division of the text into /significant components./ Disallowing segments of five characters or less would have given the best partition which can be obtained from such limited data: /By segmentation /is meant /the division of the text into / meaningful parts./ and: /By partition /we mean /the division of the text into /significant components./ Unfortunately, whether a short segment should be merged with the segment on its left or with that on its right cannot be determined.
This example suggests that, since the inventory of the symbols of the Easter Island tablets is doubtless greater than the set of punctuation and alphanumerical symbols of written English, text partitions resulting in segments of at least five symbols have very good chances of being true.- 376
Internal Structure of Meaningful Segments
It seems that much could be learned from the examination of the internal structure of the meaningful segments identified. Thus, in the exact partition of the example above, each meaningful segment is ended by either a space or a full stop. The conclusion that a space or a full stop signals the end of a meaningful segment — although not statistically justifiable on such short data — leads to the perfect segmentation of the texts into their word components.
SEGMENTATION OF THE TEXTS
Fragment A1-11 is found with considerable variations in H26-38, in P8-18, and in Q13-23. Fragment A15-35 is found in part (A15-34) in H51-80, in P33-68, and in part (A15-25) in Q40-55. Fragment A48-59 is found in H84-95, in P75-84, and in part (A56-59) in Q56-59. The position where the counterpart of A35 would be expected on tablet H is completely worn out, with glyphs H77-80 partly worn down. The absence of fragments A26-35 and A48-55 from tablet Q is consistent with Barthel's observation that tablet Q is incomplete at one end (probably broken off), since fragment Q40-55 ends line 5 and fragment Q56-59 starts line 6 of the recto of the tablet.
Thus, we have so far the correspondences:
Fragments H39-50, P19-32 and Q24-39 would be expected to correspond to A12-14, and P69-75 to A36-47. However, not only are H39-50, P19-32 and Q24-39 quite dissimilar to A12-14 (although very similar to one another), but also they are much longer (12, 14, and 16 glyphs respectively) than A12-14 (three glyphs). On the other hand, fragment P69-74, which would be expected to correspond to A36-47, is shorter (six glyphs) than A36-47 (12 glyphs). When lengths are computed by counting glyph components rather than whole glyphs fragments H39-50, P19-32 and Q24-39 are respectively 21, 20, and 23 glyphs long as against four for A12-14, and fragment P69-74 is nine glyphs long as against 19 for A36-47:
Number of: Whole glyphs Glyphs components
Fragment A36-47 is quite similar in length to H39-50, P 19-32 and Q24-39. The examination of its constituent glyphs shows a pattern similar to H39-50, P19-32 and Q24-39, in which glyph 65 occurs repeatedly where glyphs 65, 66A or 66B are found on tablets H, P, and Q, and generally precedes glyphs found in the same order on these tablets (i.e., glyphs 3, 95, 49, 1, 9, and 6).
Comparison of fragments A12-14 and P69-74 shows A12-14 to be an abridged version of P69-74:
We have then, on tablet A, a text composed of five consecutive fragments which is found on tablets H, P, and Q with the second and fourth fragments transposed, i.e.:
Therefore, the second, third, and fourth fragments of this text very probably constitute meaningful segments. On the other hand, the precise outer limits of the end fragments can only be surmised with a low degree of certainty:
We start with the longest meaningful segment identified, which is the third fragment of the text.- 378
(i) Segments A15-35, H51-80, P33-68, Q40-55
These segments exhibit a regular pattern of the type AxB, in which A is glyph 8, x is a variable element usually fused to glyph 8, and B consists of the glyph sequence 15 22f in tablets H, P, and Q, and of the single glyph 22f in tablet A. This pattern is broken in only three occurrences out of 29: fragment A23-24, which reads 301.8 22f and therefore exhibits the pattern xAB; and fragments P40 and P44, which read 81.61 instead of the expected glyph 8.61.
Glyph 81, which breaks the pattern at P40 and P44, resembles glyph 8 with the same lower end as glyph 600 (the “frigate bird”) or glyph 700, a clear representation of a fish. It is perhaps a fusion of either with glyph 8, i.e., 600:8 or 700:8, in which case we have the following correspondences:
This compares with the reversal of the pattern from AxB to xAB at A23-24 (i.e., 301.8 22f instead of an expected sequence 8 301 22f). If these are not free variations, and it seems safer at this stage to work on the hypothesis that they are not, then this segment exhibits a pattern of the type xAyB in which A consists of glyph 8, B of 22f or 15 22f, x of zero in all but two occurrences and y of zero in one occurrence only. This pattern occurs 10 times on tablets A and P and eight times or more (some glyphs are illegible through wear) on tablet H. Its third occurrence is missing on both tablets H and Q.
The frequency of repetition of this pattern, and the absence of its third occurrence from tablets H and Q, make it highly probable that each occurrence of it constitutes a meaningful segment in itself.
(ii) Segments A36-47, H39-50, P19-32, Q24-39
This is the second longest segment identified and is partly illegible on tablet H.
Glyph 74fx at A40 differs from glyph 49 at P25 and its variants at H44 and Q32, in that it lacks the protrusions at the base of the “cactus head”. Glyph 100 consists of two occurrences of glyph 22, linked by two double lines on tablet H and a double and a single line on tablet Q, and is very probably a ligature of the sequence 22 22, to which it corresponds on tablet P.
The pattern exhibited by these segments is of the type Ax, in which A consists of glyph 65, 66A or 66B and x is a variable element. It occurs precisely 10 times in A36-47 and P19-32, and probably also 10 times in H39-50, which is erased in part. It occurs either 10 or 11 times in Q24-39, depending on whether fragment Q35-37 is counted as exhibiting two occurrences of the pattern Ax (the first occurrence having a null variable element) or only one (with the doubling of the constant element A). That the two consecutive occurrences of glyph 66A in Q35-36 precede a double glyph (100) suggests that the leading element of a pattern may be optionally doubled preceding another doubled sign.
Once again the very frequency of repetition of this pattern suggests that each occurrence of it probably constitutes a meaningful segment.
(iii) Segments A12-14, H81-83, P69-74
This segment is partly erased from tablet H and completely absent from tablet Q (it may have been present on the missing end of the tablet).
As there is no discernible pattern apart from the recurrence of the same head shape on tablet P (glyphs 200, 90, and 200), this segment cannot be divided into smaller probably meaningful segments.
(iv) Leading fragment variants A1-11, H26-38, P8-18, Q13-23
The first glyph of this fragment is partly erased from tablet A, and the tops of its last three or five glyphs are erased from tablet H.
Glyphs 40A and 41A differ only in the orientation of the crescent. Wherever an anthropomorphic glyph with a head shape of the type of 220 occurs in a fragment, a similar glyph, but with a head like that of 320, can be found in some corresponding fragment. This is almost proof that the latter (call it Head 300) is a free variant of the former (call it Head 200). Note that glyphs 440 and 445 have Head 300 and glyph 695 has what appears to be a variant of Head 200 (perhaps Head 200 seen in profile). It is therefore most likely that 695 is a free variant of 445. These fragments, then, exhibit a pattern of the form ABBCCCCDDEE in which A, B, C, D, E are respectively glyphs 40A, 220.9, 440, 205, and 4.3, or variants thereof. The fact that glyph 40A (or 41A) does not occur doubled and is preceded only in tablet A by an anthropomorphic glyph suggests that it is not part of the same meaningful segment as the repeated glyphs which follow. If so, fragments A2-11, H27-38, P9-18, Q14-23 are meaningful segments, probably to be subdivided into five smaller segments each consisting of the repetition of the same glyph (single or compound) with occasional additions (e.g., 440 440′20F) or omissions (e.g., 4.3:450 4).
(v) Trailing fragment variants A48-59, H84-95, P75-84, Q56-59
Most of this fragment is missing from tablet Q. There is a good agreement among all four tablets up to glyph 1.8 (A59, H95, P84, Q59). Beyond that the text of A becomes only remotely reminiscent of the text on the other three tablets.
This segment, like A12-14 and P69-74, shows no recognisable pattern.
GRAPHIC ANALYSIS OF THREE REPEATED GLYPHS
Glyph 8 consists of a circle in a six-pointed star. It has two clearly defined variants: either all the branches of the star touch the circle (as in H66) - 381 or none does (as in P52). Call X the first variant (H66) and Y the second (P52). The occurrences of X and Y are as follows, with particular attention being paid as to whether glyph 8 is found in isolation or in ligatures:
Variant X is found in all occurrences of the glyph in tablet H, variant Y is found on tablets A and P, except as the right element of ligatures of the type a.b, where variant X is found in all cases on tablet P and apparently in free variation with variant Y on tablet A. On such stylistic evidence alone, tablets A and P would be grouped together. This tentative grouping is corroborated by a similarity of contents: tablets A and P contain a fragment 8.61 : 27B 22f (A19-20), 600? : 8.61 27A 15 22f (P40-43) which is absent from H and Q.
Glyph 22f occurs in the shape of a meniscus (M), or of a biconvex lens (B). In its biconvex lens form, barbs are found either on its left side (L) or on its right side (R), whereas in its meniscus form barbs are found on its right side only. It has thus three main variants, M, BL, and BR. These are distributed as follows:
Tablet A is characterised by the exclusive use of the B (biconvex) variant of 22f and the occasional use of its L (left-facing barbs) variant. Tablet H shows only one occurrence of the B variant and tablet Q none. Tablet P has the B and M variants occurring in roughly equal proportions (with one doubtful case at P61). There appear to be no conditioning factors for the occurrence of one variant or the other. Table A, in its exclusive use of the biconvex form, and tablet P, in its frequent use of the same form, again pair against tablets H and Q with their preponderant use of the meniscus.
Glyphs 65, 66A, 66B
Glyph 65 resembles the outline of a fish, tail up. Glyph 66A differs from it by the addition of one, and glyph 66B by the addition of two “fins”. If these are considered as variants of a same glyph, they can be characterised by their number of “fins”, i.e., 0, 1, or 2. We may further distinguish two variants of form 1, depending on whether the additional “fin” faces left (1L) or right (1R). The occurrences of these variants, again paying particular attention to the type of ligatures in which they are found, is:
Only right variants of form 1 occur on tablet P, only left variants on tablet Q. There seems to be no conditioning factor as to when form 1 or form 2 is used, only a tendency for 0, 1, or 2-“fin” variants to group in unbroken sequences of the same type, with only one exception on tablet Q, at Q28.
A Tentative Stemma from the Graphic Evidence
Let us summarise these observations in a table, with variants listed in order of decreasing frequencies or separated by a colon when occurring the same number of times:
Tablets H and Q are more similar to each other than to P in that they make more use of the X variant of glyph 8, the meniscus variant of glyph 22f and of the left-facing variant of glyph 66. Of H, P, and Q it is P which is most similar to A, in its use of the Y variant of glyph 8, and of the biconvex right-facing variant of glyph 22f. Thus, we have the stemma:
The Evidence for Two Basic Textual Structures
The comparative analysis of segments A15-35, H51-80, P33-68, Q40-55 and A36-47, H39-50, P19-32, Q24-39 has revealed a pattern in which a meaningful segment consisted of 10 successive occurrences of groups of glyphs, each consisting of a constant element (the glyph pair 8 and 22f in segment A15-35 and its parallels, and glyph 65 or its putative variants 65A and 65B in segment A36-47 and its parallels) and of a variable element. This structure is echoed in segment A2-11, and its parallels H27-38, P9-18 and Q14-23, by a pattern consisting of five successive occurrences of doubled glyphs. No such recurrent patterns are evident in segments A12-14 and A48-59 and their parallels.
A Symbolic Notation of Segment Structures
Let us at first use the notation n(Cx) to symbolise a recurrent-pattern segment, in which n is the number of times the pattern occurs, the constant element is represented by an upper case letter, the variable element by a lower case letter, and the relative order of occurrence of the symbols for the constant and variable elements is not significant. (We have seen that the variable element could occur prefixed, suffixed, infixed, or fused to the constant element.) Since the other type of segment encountered involves no repetition, and therefore no affixation of a variable element to a recurring constant element, we may simply symbolise it through a notation of the type (C), from which the repetition factor n and the variable element symbol x are absent.
In this notation, segment A15-35 and its parallels on tablets H, P, and Q may be represented by 10(Ax), segment A36-47 by 10(By), and the non- - 384 repetitive segments A12-14 and A48-59 and their parallels by (C) and (D). Segment A2-11, however, consists of five occurrences of varying glyphs sharing no constant element. However, they do share the common feature of all being doubled. This observation naturally leads to adopting a notation of the type 5(2w), expressing that segment A2-11 consists of five occurrences of the doubling of a variable element w, and forces a redefinition of the notation system proposed as n(Cx) in which n is a multiplier, C a constant element which may be a glyph or a group of glyphs or a multiplier, and x a variable element which may be a glyph or a group of glyphs. To allow for the constant element to be symbolised by a number when it is a multiplier, we have to do away with the convention that constants are symbolised by upper case letters and variables by lower case letters. This is done simply by specifying that the constant part of a group is to be given before its variable part.
Structural Transcription of the Text Fragment
Using the notation just elaborated, we have (with question marks showing doubtful points due to worn or missing fragments):
This transcription brings out two striking features of the text discussed here:
We shall restrict the conclusions to be drawn from this analysis to bringing probable answers to whether the Easter Island script constituted a true writing system or a mere mnemonic device, and to its probable history.
The Evidence for a Mnemonic Device
The Easter Island glyphs have often been likened to knots in a handkerchief, each glyph triggering the recall of a text to be chanted. This notion appears to have been prompted by the contradictory readings given by Pascuan informants, and was best summarised by Métraux (1940:404-5):
Taking into evidence all the data and the several possible explanations of - 385 the tablets, I propose the following tentative explanation. . . . The tablets were originally staves (kohau) used by the rongorongo men to beat the measure when chanting. They were decorated with carvings that became associated with chants. The symbols formed a sort of pictography in the sense that each glyph was associated with a particular sentence or group of words in a chant. The symbols did not correspond exactly to a specific chant, but each tablet could be used with many chants and several sentences were linked with each image. Since the connection between the chant and the tablet was rather loose, the signs became conventionalized and traditional. This is what Routledge's informant meant when he said that “the words were new, but the letters were old.” The magical or ornamental character of the signs superseded their pictographic value.
Métraux's conclusion is based on the following observations:
The Evidence Against a Mnemonic Device
Let us examine the four pieces of evidence above:
Thus, the hypothesis that the Easter Island script is neither a phonetic or syllabic, nor an ideographic or pictographic writing system appears to be contrary to the objective evidence, and to have been based purely on the Pascuan informants' inability to give consistent readings. That may simply be due to their having been, or having become, through years of lack of practice, functionally illiterate.- 387
Historical Implications of the Analysis
We have seen that the beginning of Tablet A contains a text also found on three other tablets. This text was subdivided into five meaningful segments occurring in the same order on all tablets, except on tablet A, where the three medial segments occur in reverse order. Fewer signs are used on tablet A to record this text than on any of the other three tablets. Finally, of those three tablets, tablet A has been shown to be most similar to tablet P both in its contents and in the writing style of the individual glyphs.
These observations suggest:
These conclusions, drawn solely from the analysis of the texts, without resorting to hypotheses as to their possible meanings, concur with the external evidence:
It is therefore probable that the Easter Island tablets represent a highly organised form of a mixed ideographic and phonetic writing system characterised by various styles of writing, reflecting perhaps local “schools”, perhaps evolution through time, which underwent at least one drastic reform brought about quite recently by the increasing scarcity of wood on the island.
Since it is unlikely that such a form of writing should evolve over a short span of time, and since it has no apparent surviving relatives outside Easter Island, it is also probable that it was elaborated to the form we know on Easter Island itself.