 |
 |
 |
|
 |
In the following chapters, we will examine the validity of theories, particularly the Catholic and the Orthodox ones (but also any other that might emerge, if none of the two can be verified). The intention is not to examine the Catholic or Orthodox theories as wholes, but rather to treat them as sets of pairs of individual conflicting theories (one Catholic/Orthodox pair for each letter or orthographic convention of the Greek script), each of which needs to be confirmed or rejected, irrespective of the validity of the other theories in the same Catholic/Orthodox set (for example, it may turn out that the Orthodox value of Ζ is better supported by the evidence, whereas the Orthodox value of Υ is not). The verdict will be based on a set of principles that will (as much as possible) conform to scientific rather than (as yet) rhetoric practice.
The notion of "science" is not equally understood by everyone, but the best definition of the work of a scientist that I have found is probably given by Harry Foundalis: "Ο επιστήμονας είναι υποχρεωμένος να παρατηρεί τον φυσικό κόσμο (μέρος του οποίου είναι και οι ανθρώπινες φυσικές γλώσσες), να συλλέγει δεδομένα, και να διατυπώνει θεωρίες που εξηγούν τα δεδομένα, που τα καλουπώνουν σε φυσικούς νόμους|The scientist is obliged to observe the natural world (part of which are the human languages), to collect data, and to formulate theories that explain the data and formalise them in natural laws".
The above definition is pretty accurate as regards most natural sciences, but may not work with a finite data size, as more than one theory may be compatible (like trying to find a circle that crosses two known points; although many circles can be ruled out, the problem still has an infinite number of solutions). Clearly, what is needed is the collection of enough data for the theory to be more robustly tested (in the example, the determination of a further point defining the circle). The revised definition defines the so-called scientific method, a definition of which is again given by Foundalis: "οτι βασιζόμαστε σε δεδομένα, όχι σε προαποφασισμένες θεωρίες· όταν η θεωρία δεν συμφωνεί με τα δεδομένα, είναι η θεωρία που καταρρίπτεται, δεν τροποποιούμε ποτέ τα δεδομένα ώστε να ταιριάξουν με τη θεωρία· οτι κάθε θεωρία πρέπει να είναι καταρριπτέα για να γίνει επιστημονικά αποδεκτή, κι οτι επιζητούμε την κατάρριψη θεωριών μέσω νέων παρατηρήσεων, όχι απαραίτητα την επιβεβαίωσή τους|that we are based on data, not on pre-conceived theories; when the theory does not agree with the data, it is the theory that is rejected, we never modify the data to match the theory; that every theory should be able to be rejected [=falsifiable] in order to be scientifically acceptable, and that we seek the rejection of theories through new observations, not necessarily their confirmation". This leads us to the first principle:
Presumed-Innocent Principle
A theory remains (potentially) valid as long as not proven invalid.
This definition may seem unnatural (particularly because it also classifies all non-falsifiable theories as valid), but the reason behind it is that, more often than not, it is not possible to positively prove the validity of a theory; for example, the statement "All swans are white" can only be proven false (by discovering a non-white swan), but cannot be positively proven, even if verified by compliant evidence (unless, of course, one can examine all swans).At first sight, this principle may seem as the legitimisation of the ad ignorantiam fallacy: "a specific belief is true because we don’t know that it isn’t true". Although the reasoning is very similar, the conclusion is different: the absence of strong contradicting data leads to the conclusion that the theory is not impossible; to conclude, instead, that the potentially true theory is certain (="true") would be the actual fallacy. In such cases, one can only seek further evidence that will either invalidate the theory or fail to do so (and keep seeking more evidence); in other words, the only way forward is to consider the negative indicators:
Incompatibility Principle
When the theory is not compatible with some data, its validity is inversely proportional to that of the contradicting data.
The reason for speaking of validity of data is that, most of the times, the data are unreliable or their interpretation uncertain. As an example, consider the aforementioned spellings ΡΗΧΣ↔ריש, ΧΣΕΝ↔שין: the data are unreliable, because I have not been able to verify that these spellings were actually there in the original;In fact, there does not seem to exist any way to verify this beyond doubt, since we only know the texts of the Bible (and all works of antiquity) through much later manuscripts, which may not be faithful copies of the originals, but often edited versions thereof. furthermore, the interpretation of the data is ambiguous, since it is not clear whether the use of the digraph ΧΣ by the Septuagint to represent the Hebrew letter ש (=[ʃ], according to the current pronunciation) corresponds to an established value of the digraph as used in the Old-Attic orthographic system or is a new convention established after IV BC. Such an isolated questionable piece of evidence can obviously not bear enough weight to invalidate the theory that ΧΣ=Ξ=[ks], but it does cast some doubt on its validity.
When the collected data cannot be safely considered accurate or trustworthy, the theory should not be expected to 100% match all data, but to describe them imprecisely. Enter statistical hypothesis testing, which calls for collection of even more data for compensating for the uncertainty of the data. The higher the percentage of (uncertain) data that match a theory the more likely it is that the theory is valid.
Aggregation Principle
The probability of a theory being valid is considerable, only if supporting data of uncertain validity are statistically significant.
It is admitted that the above definition is somewhat vague, since the notion of "statistical significance" is not precisely specified. Unfortunately, a decision about the amount which may be considered significant will have to be taken ad hoc, depending on the nature of the data; for example, considering Allen's aforementioned examples of initial voicing of Greek words borrowed into Latin, the four listed examplesAllen insinuates, with the expression "and so on", that there are numerous further examples, which are (obviously for reasons of economy) not mentioned, but neither can I think of nor have I read about any other similar example (apart from a couple more cited by Sturtevant). are certainly statistically insignificant, considering that random sound change during borrowing is not unheard of.
The objective dealt with in this Website relates to a particular field of science, which is concerned with the reconstruction of some part of the past (in this case, the evolution of Attic phonology). The "reconstructionist" has a far harder task that most other scientists, for instance the physicists who study the (perennial) laws of nature. Although they both can formulate theories based on observed data, the latter can generate further data (by devising and carrying out controlled experiments and/or by further observing a still-occurring phenomenon) that would better test the theory, whereas the former cannot (for the simple reason that the observed period is... gone); once all evidence (e.g., inscriptions, testimonies, internal evidence, etc., in the case of reconstructing the phonology of a language of the past) has been collected, there is no or very little possibility for further evidence that will verify a theoryAs a matter of fact, the last 100 years or so have not brought forward any decidedly important further evidence on ancient-Greek pronunciation., whereas in experimental or observational science (such as the study of physical laws) further, essentially eternal, experimentation or observation is always possible. Thus, the "reconstructionist" has to rely on whatever traces or remnants have been left by the object of study and can only hope that more data will come to light as test samples for the theories. Unfortunately, most of the times this does not happen and the available data are insufficient for confirming or invalidating the theories. Without access to direct evidence or contemporary reference points, the accurate reconstruction of the past appears to be an impossible task; however, this has not deterred the "reconstructionists", who have not faltered in front of far harder tasks: they have devised several tricks to fill the gaps and they can now tell us how the Universe was created, how a unicellular organism mutated into the unfathomably complex humans and what language was spoken by Adam's (or Lucy's) close descendants. Intriguing as all these stories may be, it is seldom made clear that they contain a large amount of speculation.
In fact, reconstruction of the past often oscillates between science and fiction. I was watching a documentary (#10, "River of Death") on History Channel about the struggle of Albertosaurus vs Pachyrhinosaurus. A scientist was explaining (watch here) that Albertosaurus’s behaviour was inferred by that of modern pack animals. Another scientist noted that "hunting behavior does not fossilize", which only means that "possible hunting behavior [is] resting on so many inferences and hunches that we can't possibly know what actually occurred; active hunting behavior doesn't fossilize" (as explained in another online article). To present as fact something that is not based on actual evidence, but rather on our (limited) experience from seemingly similar situations, is (at least in my eyes) next to fraud, unless one explicitly states that the reconstruction is based mainly on the artist's imagination and should be regarded merely as a work of art. In that case, it would be more honest to simply say "we don't know for sure" (but then there would be no documentary and no money making), rather than ask "what else could it be, if not what we already know?" (in many instances, reality has turned out to exceed our experience or even imagination). The least we can do is state the following principle.
Extrapolation Principle
Inferences from the present about the past can only indicate possibilities and do not constitute proofs.
Another trick to "fill the gaps" is to make "reasonable" assumptions about the parts that we do not know. One such example is the argument that having too many graphemes for the same ([i]) sound in a language is a luxury, therefore classical Greeks must have pronounced each one of ι, η, υ, ει, οι, υι, ῃ differently. This is a very reasonable assumption, but the projection of this reasoning to the present leads to the surprising conclusion that neither do modern Greeks pronounce them equally, which is certainly not true (a typical example of reductio ad absurdum).
Invariability Principle
Assumptions about the past, even if they seem reasonable, are not sound, if they do not hold true in the present.
Often, a counterargument in the case of the above example is that the two (classical and modern Greece) are not comparable, e.g., because modern Greeks have a long literary tradition behind them, which they cannot ignore, whereas classical Greeks did not (which is not true considering, e.g., the continuous teaching of the Homeric poems for more than three centuries before the classical age). Although the argument in that specific case is faulty, in general, application of the Invariability Principle will be conditional on a verification that the circumstances are... invariant, namely that the same circumstances are prevalent in both the past and in the present.
The previously described list of fallacies, which are put forward by (those that I call) the "great masters of (linguistic) reconstruction" and served as the basis for my questionable "proof" about the value of ancient Ξ and Ψ, makes it clear that reconstructive linguistics is in dire need of solid proof methods. The obvious choice to fulfill this need is to look to the direction of Mathematics.
The usual way to prove a statement in Mathematics is to start from a number of premises, upon which some form of logic is applied to arrive at the conclusion, namely the statement to be proved. As an example, consider the following mathematical formulas and corresponding examples:
| (A = B)|A equals B ∧|and (B = C)|B equals C ∴|therefore A = C|A equals C | (A ∈ B)|A belongs to (set) B ∧|and (B ⊂ C)|B is a subset of C ∴|therefore A ∈ C|A belongs to C | |
| Premises | (P11) The prettiest island is Santorini | (P21) The prettiest island is in Greece |
| (P12) Santorini is Greek | (P22) Greece is part of Europe | |
| Conclusion | (C1) The prettiest island is Greek | (C2) The prettiest island is in Europe |
There are four possible types of premises:
It is more difficult to categorize the different rules of logic usually applied on the premises. The most common kind is deduction (ἐπαγωγή), wherein a common link between two premises establishes the relationship between the other parts of the premises, as in the above examples.
Another known possibility is contradiction, wherein the negation (contradiction) of the conclusion is shown to have a consequence that is demonstrably false. For example: John does not go to church; therefore, John is not a devout Christian, because if he were, he would often go to church. However, proof by contradiction may be considered as merely another form of deduction.
| (A = B)|A equals B ∧|and (C = ¬B ↔ B = ¬C)|C equals the opposite of B; equivalently, B equals the opposite of C ∴|therefore A = ¬C|A equals the opposite of C | |
| Premises | (P31) John is not a church goer |
| (P32) All devout Christians are regular church goers | |
| Conclusion | (C3) John is not a devout Christian |
We have seen that deduction is the usual scientific method for proving something. It is a relatively simple method to follow, but it is often disregarded, particularly in the circles of the so-called human sciences, as we have seen before. The reason has been very elegantly explained by Armen Zemanian in an inspired 1994 paper of his (ZEMA94): "The usual techniques for proving things are often inadequate because they are merely concerned with truth"! He then goes ahead and provides "an (undoubtedly incomplete) list" of alternative proof methods, such as the
An augmented version of this list can also be found online, of which my favourite (as we will often encounter this particular type of "proof" in the course of examining "scientific" proofs about ancient-Greek pronunciation) is:
These kinds of arguments (or "proofs"), which are very often put forward by "reconstructionists", are patently invalid and constitute fallacies. Fallacy is formally defined as "a failure in reasoning which renders an argument invalid".Note that fallacy invalidates the argument (or proof) and the certainty about the conclusion, but not necessarily the conclusion per se; it is perfectly possible that a correct conclusion is based on false reasoning (e.g., Rome is Greek; Greece is in Europe; therefore, Rome is in Europe), but its "correctness" has to be established based on sound premises (i.e., that Rome is Italian and Italy is in Europe) and reasoning. One can find many lists of logical fallacies and it is not essential at this point to go through any of them (we will refer to them on the spot, namely as we encounter them while examining a particular argumentation), but to identify potential sources of fallacy.
In order for the proof to be sound (and the conclusion valid), both the premises and the applied logic should be well-founded. Considering the former, the different kinds of premises defined above have different "predisposition" to fallacy:
Sometimes the fallacy lies in the logic. Consider, for example, the following proofs:
| Premises | (P41) Today is Monday | (P51) Angelina's father is Jon |
| (P42) Mondays are boring | (P52) Jon's last name is Voight | |
| Conclusion | (C4) It will rain tonight | (C5) Angelina's last name is Voight |
(C4) completely disregards the rules of logic, as the only reasonable conclusion that can be derived from the premises is that "today will be boring"; instead, it is a typical example of jumping to unwarranted conclusions. Its verisimilitude relies on a common prejudice, namely that boredom is often caused by bad weather, which, however, neither is part of the premises (either as an axiom or a fact) nor would it necessarily lead to this particular conclusion, if it were.
(C5) seems reasonable, but we know that this is no longer the case; hence, there must be something wrong with the reasoning. Indeed, the missing link is obviously the assumption that children get their father's last name and keep it for life. This kind of assumption, not featuring in the logic or among the premises, is known as a hidden premise. Hidden premises are sometime hard to spot, but they must be identified and examined more closely. In this particular case, the identified hidden premise is not correct (Pablo was another example, although not officially). It might also be considered that the faulty logic in the other example is due to a (fallacious) hidden premise, i.e., that on boring nights it rains.
As an example of how the above can be put into practice, let's (re)consider the initial argument on Latin C (and the assertion that VICI was pronounced by Caesar as [ˈwiːki]), which can be broken down as follows:
| "in transliterated Greek words, Latin C represents Greek Κ ("Cerberus", "cynicus", etc), which has always been pronounced [k]; ergo C=[k]" | |
| Premises | (P1) Latin C (almost always) represents Greek Κ in transliterated Greek words |
| (P2) Greek Κ was always (i.e., in all environments) pronounced [k] | |
| Conclusion | (C) Latin C was always (i.e., in all environments) pronounced [k] |
(P1) is an established fact and can be easily verified from a variety of sources. However, (P2) is a conjecture, since we have no direct (=audio) evidence and since it has not been established (in the argument) sufficiently to serve as a solid premise for the proof.It is not clear how we can be sure about a uniform value of Greek Κ in ancient times. If we judge based on the modern pronunciation (as well as most modern languages), we have to accept at least the possibility of a palatal allophone before front vowels. If we rely on ancient testimonies (e.g., the so-called "grammarians"), we should not expect the ancients to have had a better feeling of the language than the average modern speaker (who cannot tell the difference between the Κs of, e.g., κοίτα|look!=[ˈcita] and κότα|hen, chicken=[ˈkota]). If we look at graphic evidence, it would be difficult to explain, e.g., the use of Ϙ in the place of Κ before (some) back vowels without the existence of allophones of Κ. Even if we accept (P2) as an axiom, the conclusion (C) does not follow from these two premises. Although it is not easy to spot the fault in the logic, if one looks carefully, one will realise that what appears as a "link" between the two premises ("Greek K") is actually not such: in (P1) it relates to a written Κ and in (P2) to a spoken Κ. The argument relies actually on the hidden premise that:
| Premises | (P1) Latin C (almost always) represents Greek Κ in transliterated Greek words |
| (P2) Greek Κ was always (i.e., in all environments) pronounced [k] | |
| (P3) The intention in transliterations, particularly the consistent ones, is to faithfully represent the sound of the original language | |
| Conclusion | (C) Latin C was always (i.e., in all environments) pronounced [k] |
Only if (P3) is adopted as a premise is the logic sound and the conclusion correct. However, not only is this claim impossible to prove (as it relates to psychology and, hence, subjective factors), but it is also contradicted by our experience from modern transliterations. Consider, for example, the case of German CH, which has two allophones: a velar ([x]) after back vowels and a palatal ([ç]) after front vowels and consonants. German CH is consistently transliterated in Greek with Χ (Bach↔Μπαχ, Reich↔Ράιχ, Münchausen↔Μυνχάουζεν, Aachen↔Άαχεν). The situation in Greek is, however, different: Χ does have the same two allophones ([x] and [ç]), but they are determined by the following vowel, namely velar ([x]) before back vowels and consonants and palatal ([ç]) before front vowels. This does not prevent the Greek transliteration to follow the German orthography, even in cases where it is possible to represent the exact German pronunciation (Michael↔Μίχαελ not Μίχιαελ, Richard↔Ρίχαρντ not Ρίχιαρντ, etc). Thus, it is clear that, in this example, the principle set out in (P3) is not followed, but the transliteration rather abides to orthography; in other words, in accordance with the Invariability Principle, (P3) is not a sound assumption.
All in all, only one out of the three premises supporting the "proof" is certain, the other two being unlikely (P2) and occasionally (i.e., not universally) true (P3). The argument is, therefore, invalid and the conclusion (C) is not justified (by this particular argument).
Having disqualified the argument as a definite proof, it would also be meaningful to investigate whether the conclusion (C) may at least serve as the only theory that would explain the available data, namely the consistent transliteration C↔Κ (represented by premise (P1) above). After all, the above analysis shows that (C) is not an absolute certainty, but it is not ruled out either; in other words, (C) is still a possible explanation for (P1). There are, however, several candidate theories that might do the trick (since the value [k] for both C and Κ before back vowels and consonants seems all but certain and is not disputed, the following theories refer to their values before front vowels):
Theory (C) is the above "conclusion", that is predominantly supported by most Latin scholars; (C') is its adjustment to the present value of Greek Κ; (C'') reflects, more or less, the present situation as represented by modern Greek and modern Italian, respectively. This list could be expanded with innumerable further possible values for each of the two letters, but it is better to be realistic and restrict the possibilities to the values suggested by the surviving tradition ([c] for Greek and [t͡ʃ] for, e.g., Italian and Romanian).
Clearly, (C) and (C') are compatible with (i.e., not contradicted by) the data: if two letters represent identical values, it is only reasonable that one is used in lieu of the other when transliterating. Even though (C'') might at first seem implausible in light of the data, one has to consider what a transliterator would do, had this been the case (i.e. that Κ before front vowels were a velar or palatal and C an affricate). In that case, Greek would lack any suitable representation of affricate sibilants and Latin would lack a voiceless velar sound in an environment before front vowels.
If a Greek were to use a letter different from Κ for representing the (assumed) affricate sound of C, then Σ and (possibly) Ζ (the sibilant monophthongs) would be the only reasonable alternatives. However, Ζ was most certainly a voiced sound (as it is today) and would not be "closer" to [t͡ʃ] (or any other voiceless affricate value of C) than would be Κ, whereas Σ had already a Latin counterpart (S) and its use also for C would lead to unnecessary confusions (for the same reason, Ξ, the counterpart of Latin X, would not be appropriate either and Ψ would be too front a sound for representing a coronal affricate). The present practice of using ΤΣΕ, ΤΣΙ for the Italian CE, CI was established much later than antiquity (when the newly-introduced CHE, CHI were contrasted to the existing CE, CI in Italian and when the affricates were made part of Greek phonology), the digraph <ΤΣ> being foreign to Greek orthography (and most likely the affricate [t͡s] being foreign to Greek pronunciation, since dentals were, in general, dropped before sibilants). Thus, the use of Κ (particularly if it had the palatal value [c]) would be the best option.
For transliteration from Greek to Latin, a Roman would have no way to render the syllables [ke], [ki] (or [ce], [ci]). K was an almost foreign letter in Classical Latin and, in the few cases it was used, it is generally found before A (the reason for that is not clear). The present convention of Spanish and French to render these syllables as QUE and QUI would not be applicable, as <QV> (note that the grapheme <U> did not exist) was rather a labialised velar ([kʷ]) in classical Latin (which developed to [ke], [ki] or [ce], [ci] in Romance). The Italian and Romanian modern renderings CHI and CHE are of unknown (at least to me) origin and would be also inapplicable, since CH was the transliteration of Greek Χ.
One should also not forget the possibility that the transliteration did not follow pronunciation, but orthography, for two letters that had the same value ([k]) in the majority of environments, in a manner analogous with the correspondence CH↔Χ between German and Greek. Any phonetic difference between the two letters (always before front vowels) could be either imperceptible for untrained speakers (as is, e.g., between the word-initial voiceless plosives, which are aspirated in English and unaspirated in French, Greek, etc, but are perceived as practically the "same" by normal speakers of these languages) or construed as a regional peculiarity (as are, e.g., the affricated and palatalised variants of Κ in Cretan and mainstream Greek, respectively).
To summarise, even though (C) and (C') fair better with the data (no further assumptions have to be made), (C'') is also reasonably compatible with the data. In accordance with the Presumed-Innocent Principle, all three theories must be considered valid, as far as the particular evidence (transliteration C↔Κ) is concerned.
As (P1) does not serve to disqualify any of the candidate theories, we need to seek further evidence and check the theories against them. For example, the following facts:
make one wonder:
The answers to the above questions are obvious, namely that there is a very high probability that (P4) is also applicable to Latin and that it is more plausible that the language brought by veterans and administrators to the new lands already comprised an affricate C. Therefore, in accordance with the Incompatibility Principle, one has to conclude that theory (C) is almost certainly false (given the universality of velar "fronting" before front vowels) and that theory (C') is rather implausible (given the identical development in the various geographically dispersed Romance languages).
Of course, this analysis is incomplete, since only three pieces of evidence are examined. The three theories have to be tested against all available evidence before they are ranked in terms of their credibility.However, there do not seem to be any substantial further evidence. Allen, for one, cites the evidence of transliterations between Latin on one side and Greek or (the later ones of) Celtic and Germanic on the other, an uncertain alliteration by Livy, an indirect statement by Varro about the "velar value of n" in anceps and the lack of "suggestion of anything other than a velar plosive" by the grammarians, as evidence for a velar value of C in all environments, and some internal evidence (which he doubts in the second edition!) about the conversion of E "followed by a 'dark' l" to O or V (=u) in addition to the already mentioned allographic use of C, K, Q before I/E, A, O/V respectively and the allophonic nature of [k] in English, as evidence of a somewhat different value of C before front vowels (ALLE78, pp. 14-15).
It is evident that the discussion so far, leads to the conclusion that the traditionally accepted value among the Latin scholars (i.e., (C) or (C') or something in between) may not be correct. So, another argument one may put forward against (C'') could be roughly formulated as "Who are you to doubt renown linguists like Allen?". This is a combination of an ad hominem attack ("you are nobody, so you are certainly wrong") and an argument from authority ("if Allen says so, it must be true"). However, this is exactly the kind of religious stance that I have been... preaching against since the first chapter. The question is not to adopt someone else's creed, but to judge for oneself which creed appears more... credible. Arguments of this type will henceforth receive no attention.
There is one final step before proceeding to the examination of the arguments for the phonetic values of individual Greek letters (and combinations thereof).
|
|
|
|
|