Mass comparison is a method developed by Joseph Greenberg to determine the level of genetic relatedness between languages. It is now usually called multilateral comparison. The method is generally rejected by linguists (Campbell 2001, p. 45), though it has some supporters.
In spite of widespread skepticism about his method, some of the relationships established by Greenberg gradually came to be generally accepted (e.g. Afro-Asiatic and Niger–Congo). Others are widely accepted though disputed by some (e.g. Nilo-Saharan), others are predominantly rejected but have some defenders (e.g. Khoisan), while others continue to be widely rejected and have only a handful of defenders (e.g. Amerind).
The application of mass comparison led Greenberg not only to propose novel classifications but to break apart previously accepted ones. The best-known example is his rejection of the Hamitic language family.
Contents |
Mass comparison involves setting up a table of basic vocabulary items and their forms in the languages to be compared. The table can also include common morphemes. The following table was used by Greenberg (1957, p. 41) to illustrate the technique. It shows the forms of six items of basic vocabulary in nine different languages, identified by letters.
| A | B | C | D | E | F | G | H | I | |
|---|---|---|---|---|---|---|---|---|---|
| Head | kar | kar | se | kal | tu | tu | to | fi | pi |
| Eye | min | ku | min | miŋ | min | aš | min | idi | iri |
| Nose | tor | tör | ni | tol | was | waš | was | ik | am |
| One | mit | kan | kan | kaŋ | ha | kan | kεn | he | čak |
| Two | ni | ta | ne | kil | ne | ni | ne | gum | gun |
| Blood | kur | sem | sem | šam | i | sem | sem | fik | pix |
The basic relationships can be determined without any experience in the case of languages that are fairly closely related. Knowing a bit about probable paths of sound change allows one to go farther faster. An experienced typologist — Greenberg was a pioneer in the field — can quickly recognize or reject several potential cognates in this table as probable or improbable. For example, the path p > f is extremely frequent, the path f > p much less so, enabling one to hypothesize that fi : pi and fik : pix are indeed related and go back to protoforms *pi and *pik/x, while knowledge that k > x is extremely frequent, x > k much less so enables one to choose *pik over *pix. Thus, while mass comparison does not attempt to produce reconstructions of protolanguages — according to Greenberg (2005:318) these belong to a later phase of study — phonological considerations come into play from the very beginning.
The tables used in actual research involve much larger numbers of items and languages. The items included may be either lexical, such as 'hand', 'sky', and 'go', or morphological, such as PLURAL and MASCULINE (Ruhlen 1987, p. 120).
Critics of mass comparison generally assume that mass comparison has no means to distinguish borrowed forms from inherited ones, unlike comparative reconstruction, which is able to do so through regular sound correspondences. These questions were addressed by Greenberg (1957, p. 39) as of the 1950s. According to him, the key points are as follows :
Greenberg considered that the results achieved through this method approached certainty (Greenberg 1957, p. 39): "The presence of fundamental vocabulary resemblances and resemblances in items with grammatical function, particularly if recurrent through a number of languages, is a sure indication of genetic relationship."
It is often reported that Greenberg sought to replace the comparative method with a new method, mass comparison (or, among his less scrupulous critics, "mass lexical comparison"). He consistently rejected this characterization, stating for instance, "The methods outlined here do not conflict in any fashion with the traditional comparative method" (1957:44) and expressing wonderment at "the strange and widely disseminated notion that I seek to replace the comparative method with a new and strange invention of my own" (2002:2). According to Greenberg, mass comparison is the necessary "first step" in the comparative method (1957:44), and "once we have a well-established stock I go about comparing and reconstructing just like anyone else, as can be seen in my various contributions to historical linguistics" (1990, quoted in Ruhlen 1994:285). Reflecting the methodological empiricism also present in his typological work, he viewed facts as of greater weight than their interpretations, stating (1957:45):
The thesis of mass comparison, then, is that:
It is widely believed among linguists that Greenberg's method of language classification was limited to comparisons of words alone, to the neglect of grammatical elements, which could often provide more decisive evidence for language relationship or non-relationship. Greenberg's method, many of them state, is known as "mass lexical comparison" (e.g. Kessler 2001, p. ix).
In reality, Greenberg never used the phrase "mass lexical comparison". It is not found in any of his principal works on language classification or anywhere else in his works. It does not occur, for instance, in Greenberg 1955, 1957, 1960, 1971, 1987, 2000–2002, or 2005. Furthermore, the phrase "mass lexical comparison" is not used by any of Greenberg's supporters, for example John Bengtson, Harold C. Fleming, Paul Newman, Merritt Ruhlen, Timothy Usher, or William S.-Y. Wang.
Probably the most influential critique of mass comparison is that by Lyle Campbell. According to Campbell, Greenberg's method of language classification "relies on inspectional similarities in vocabulary alone" (Campbell 2004, p. 348):
Greenberg consistently rejected the assertion that he relied on lexical comparisons alone. He went so far as to publish the first volume of his Eurasiatic work, devoted to grammatical comparisons, two years before the second volume, devoted to vocabulary comparisons, in an attempt to get the point across (2000: vii):
Typical of Greenberg's use of morphological comparisons is this (1955: 82, 83):
Throughout his work on African classifications, Greenberg's method is the same: he begins with a discussion of morphology, and then follows this up with a few pages of lexical comparisons.
Comparisons of the number of pages devoted to grammatical versus lexical comparisons from the beginning, middle, and end of Greenberg's career show the same pattern: In Greenberg 1955 (Africa), there are 38 pages of grammatical comparisons, versus 25 of lexical comparisons. In Greenberg 1971 (Indo-Pacific), there are 21 pages of grammatical comparisons versus 34 of lexical comparisons. In Greenberg 2000-2002 (Eurasiatic), there are 179 pages of grammatical comparisons versus 181 of lexical comparisons.
Greenberg did not come by this position late in his career. As early as 1956, he stated (2005: 60):
In sum, mainstream historical linguists believe that Joseph Greenberg advocated and practiced a technique limited to lexical comparison. Greenberg's published writings show that he advocated the use of both lexical and grammatical data and that he carried out this theoretical desideratum in practice. No such technique as "mass lexical comparison" has ever existed.
The conflict over mass comparison can be seen as a dispute over the legacy of the comparative method, developed in the 19th century, primarily by Danish and German linguists, in the study of Indo-European languages.
Since the development of comparative linguistics in the 19th century, a linguist who claims that two languages are related, whether or not there exists historical evidence, is expected to back up that claim by presenting general rules that describe the differences between their lexicons, morphologies, and grammars. The procedure is described in detail in the comparative method article.
For instance, one could prove that Spanish is related to Italian by showing that many words of the former can be mapped to corresponding words of the latter by a relatively small set of replacement rules—such as the correspondence of initial es- and -s, final -os and -i, etc. Many similar correspondences exist between the grammars of the two languages. Since those systematic correspondences are extremely unlikely to be random coincidences, the most likely explanation by far is that the two languages have evolved from a single ancestral tongue (Latin, in this case).
All pre-historical language groupings that are widely accepted today—such as the Indo-European, Uralic, Algonquian, and Bantu families—have been proved in this way.
The actual development of the comparative method was a more gradual process than Greenberg's detractors suppose. It has three decisive moments. The first was Rasmus Rask's observation in 1818 of a possible regular sound change in Germanic consonants. The second was Jacob Grimm's extension of this observation into a general principle (Grimm's law) in 1822. The third was Karl Verner's resolution of an irregularity in this sound change (Verner's law) in 1875. Only in 1861 did August Schleicher, for the first time, present systematic reconstructions of Indo-European proto-forms (Lehmann 1993:26). Schleicher, however, viewed these reconstructions as extremely tentative (1874:8). He never claimed that they proved the existence of the Indo-European family, which he accepted as a given from previous research — primarily that of Franz Bopp, his great predecessor in Indo-European studies.
Karl Brugmann, who succeeded Schleicher as the leading authority on Indo-European, and the other Neogrammarians of the late 19th century, distilled the work of these scholars into the famous (if often disputed) principle that "every sound change, insofar as it occurs automatically, takes place according to laws that admit of no exception" (Brugmann 1878).[1]
The Neogrammarians did not, however, regard regular sound correspondences or comparative reconstructions as relevant to the proof of genetic relationship between languages. In fact, they made almost no statements on how languages are to be classified (Greenberg 2005:158). The only Neogrammarian to deal with this question was Berthold Delbrück, Brugmann’s collaborator on the Grundriß der vergleichenden Grammatik der indogermanischen Sprachen (Greenberg 2005:158-159, 288). According to Delbrück (1904:121-122, quoted in Greenberg 2005:159), Bopp had proved the existence of Indo-European in the following way:
Furthermore, Delbrück took the position later enunciated by Greenberg on the priority of etymologies to sound laws (1884:47, quoted in Greenberg 2005:288): "obvious etymologies are the material from which sound laws are drawn."
The opinion that sound correspondences or, in another version of the opinion, reconstruction of a proto-language are necessary to show relationship between languages thus dates from the 20th, not the 19th century, and was never a position of the Neogrammarians. Indo-European was recognized by scholars such as William Jones (1786) and Franz Bopp (1816) long before the development of the comparative method.
Furthermore, Indo-European was not the first language family to be recognized by students of language. Semitic had been recognized by European scholars in the 17th century, Finno-Ugric in the 18th. Dravidian was recognized in the mid-19th century by Robert Caldwell (1856), well before the publication of Schleicher's comparative reconstructions.
Finally, the supposition that all of the language families generally accepted by linguists today have been proved by the comparative method is untrue. For example, although Eskimo–Aleut has long been accepted as a valid family, "Proto-Eskimo–Aleut has not yet been reconstructed" (Bomhard 2008:209). Other families were accepted for decades before comparative reconstructions of them were put forward, for example Afro-Asiatic and Sino-Tibetan. Many languages are generally accepted as belonging to a language family even though no comparative reconstruction exists, often because the languages are only attested in fragmentary form, such as the Anatolian language Lydian (Greenberg 2005:161). Conversely, detailed comparative reconstructions exist for some language families which nonetheless remain controversial, such as Altaic and Nostratic (however, a specification is needed here: Nostratic is a proposed proto-proto-language, while Altaic is a "simple" proto-language - with Altaic languages widely accepted as typologically related. Detractors of both proposals simply claim that the data collected to prove by comparativism the existence of both families is scarce, wrong and non sufficient. Keep in mind that regular phonological correspondences need thousands of lexicon lists to be prepared and compared before being established. These lists are lacking for both the proposed families. Furthermore, other specific problems affect "comparative" lists of both proposals, like the late attestation for Altaic languages, or the comparison of not certain proto-forms, like proto-Kartvelian, for Nostratic.).[1][2]
Greenberg claimed that he was at bottom merely continuing the simple but effective method of language classification that had resulted in the discovery of numerous language families prior to the elaboration of the comparative method (1955:1-2, 2005:75) and that had continued to do so thereafter, as in the classification of Hittite as Indo-European in 1917 (Greenberg 2005:160-161). This method consists in essentially two things: resemblances in basic vocabulary and resemblances in inflectional morphemes. If mass comparison differs from it in any obvious way, it would seem to be in the theoretization of an approach that had previously been applied in a relatively ad hoc manner and in the following additions:
The positions of Greenberg and his critics therefore appear to provide a starkly contrasted alternative:
Besides systematic changes, languages are also subject to random mutations (such as borrowings from other languages, irregular inflections, compounding, and abbreviation) that affect one word at a time, or small subsets of words. For example, Spanish perro (dog), which does not come from Latin, cannot be rule-mapped to its Italian equivalent cane (the Spanish word can would be the Latin-derived equivalent but is much less used in everyday conversations, being reserved for more formal purposes). As those sporadic changes accumulate, they will increasingly obscure the systematic ones — just as enough dirt and scratches on a photograph will eventually make the face unrecognizable.
On this point, Greenberg and his critics agree, as over against the Moscow school, but they draw contrasting conclusions:
An unexpected offshoot of the controversy over mass comparison has been a new debate over the mathematics of language classification.
From an early date, Greenberg (esp. 1957:36-44) argued that mass comparison rests on a mathematical basis. Although "[t]he most straightforward method of eliminating chance would be the calculation of the expected number of chance resemblances between two languages", "[i]n practice, this proves extremely difficult" (1957:37). For one thing, "it requires ... a frequency weighting of phonemes", along with an evaluation of "the possibilities of phonemic combination" (ib.). Complicating the issue, it is often the case that languages with very different levels of relationship show "approximately the same" number of resemblances (ib.). The solution, according to Greenberg, is to consider multiple languages at once (ib.): the key lies in their overlapping relationships, not in their two-by-two resemblances. According to him (1957:38-39):
In a series of articles, Donald Ringe (1992, 1993, 1995, 1996) has attacked Amerind, Nostratic, and other proposals of "long-range" language classification on mathematical grounds. In his view "the remoter relationships cannot be demonstrated because the languages in question have diverged too much" (ib.). According to Ringe, sound change renders linguistic relationships untraceable over time. The level of chance resemblances between languages, however, is relatively high. Not having considered the possibility of chance resemblances, the relationships Greenberg alleges are indistinguishable from coincidence.
Greenberg replied that he had considered this very possibility in his previous work, which Ringe had failed to cite and thus, presumably, to read (Greenberg 1993, p. 79). This began an ongoing series of exchanges between scholars inclined to support mass comparison and those inclined to oppose it.
Today there are three distinct views on the mathematics of language classification with respect to mass comparison.
In spite of the apparently intractable nature of the conflict between Greenberg and his critics, a few linguists have begun to argue for its resolution. Edward Vajda, acclaimed for his recent demonstration of Dené–Yeniseian, attempts to stake out a position that is sympathetic to both Greenberg’s approach and that of its critics, such as Lyle Campbell and Johanna Nichols.[3] George Starostin, a member of the Moscow school, argues that Greenberg’s work, while perhaps not going beyond inspection, presents interesting sets of forms that call for further scrutiny by comparative reconstruction, specifically with regard to the proposed Khoisan [4] and Amerind [5] families.
Anti-Greenbergian
Greenbergian