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. Mass comparison is rejected by most linguists, and its continued use is primarily restricted to fringe linguistics.
Some of the top-level relationships Greenberg named are now generally accepted thanks to analysis with other, more widely accepted linguistic techniques, though they had already been posited by others (e.g. Afro-Asiatic and Niger–Congo). Others are accepted by many though disputed by some prominent specialists (e.g. Nilo-Saharan), while others are almost universally rejected (e.g. Eurasiatic, Khoisan and Amerind).
Methodology
The idea of mass comparison method is that a group of languages is related when they show numerous resemblances in vocabulary, including pronouns, and morphemes, forming an interlocking pattern common to the group. Unlike the comparative method, mass comparison does not require any regular or systematic correspondences between the languages compared; all that is required is an impressionistic feeling of similarity. Greenberg does not establish a clear standard for determining relatedness; he does not set a standard for what he considers a "resemblance" or how many resemblances are needed to prove relationship.
Mass comparison is done by setting up a table of basic vocabulary items and their forms in the languages to be compared for resemblances. The table can also include common morphemes. The following table was used by Greenberg to illustrate the technique. It shows the forms of six items of basic vocabulary in nine different languages, identified by letters.
ABCDEFGHIHead kar kar se kal tu tu to fi piEye min ku min miŋ min aš min idi iriNose tor tör ni tol was waš was ik amOne mit kan kan kaŋ ha kan kεn he čakTwo ni ta ne kil ne ni ne gum gunBlood kur sem sem šam i sem sem fik pix
According to Greenberg, basic relationships can be determined without any experience in the case of languages that are fairly closely related, though knowledge of probable paths of sound change acquired through typology allows one to go farther faster. For instance, the path p > f is extremely frequent, but the path f > p is much less so, enabling one to hypothesize that fi : pi and fik : pix are indeed related and go back to protoform *pi and *pik/x. Similarly, while knowledge that k > x is extremely frequent, x > k is much less so enables one to choose *pik over *pix. Thus, according to Greenberg (2005:318), phonological considerations come into play from the very beginning, even though mass comparison does not attempt to produce reconstructions of protolanguages as these belong to a later phase of study. The tables used in actual mass comparison 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. For Greenberg, the results achieved through mass comparison approached certainty: "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."
Relation to the comparative method
As a tool for identifying genetic relationships between languages, mass comparison is an alternative to the comparative method. Proponents of mass comparison, such as Greenberg, claim that the comparative method is unnecessary to identify genetic relationships; furthermore, they claim that it can only be used once relationships are identified using mass comparison, making mass comparison the "first step" in determining relationships (1957:44). This contrasts with mainstream comparative linguistics, which relies on the comparative method to aid in identifying genetic relationships; specifically, it involves comparing data from two or more languages. If sets of recurrent sound correspondences are found, the languages are most likely related; if further investigation confirms the potential relationship, reconstructed ancestral forms can be set up using the collated sound correspondences.
However, Greenberg did not entirely disavow the comparative method; he stated that "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) and accused mainstream linguists of spreading "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). Earlier in his career, before he fully developed mass comparison, he even stated that his methodology did not "conflict in any fashion with the traditional comparative method" (1957:44). However, Greenberg sees the comparative method as playing no role in determining relationships, significantly reducing its importance compared to traditional methods of linguistic comparison. In effect, his approach of mass comparison sidelined the comparative method with a "new and strange invention of his own".
Reflecting the methodological empiricism also present in his typological work, he viewed facts as of greater weight than their interpretations, stating (1957:45):
[R]econstruction of an original sound system has the status of an explanatory theory to account for etymologies already strong on other grounds. Between the *vaida of Bopp and the *γwoidxe of Sturtevant lie more than a hundred years of the intensive development of Indo-European phonological reconstruction. What has remained constant has been the validity of the etymologic relationship among Sanskrit veda, Greek woida, Gothic wita, all meaning "I know", and many other unshakable etymologies both of root and of non-root morphemes recognized at the outset. And who will be bold enough to conjecture from what original the Indo-Europeanist one hundred years from now will derive these same forms?
Criticism
Errors in application
The presence of frequent errors in Greenberg's data has been pointed out by linguists such as Lyle Campbell and Alexander Vovin, who see it as fatally undermining Greenberg's attempt to demonstrate the reliability of mass comparison. Campbell notes in his discussion of Greenberg's Amerind proposal that "nearly every specialist finds extensive distortions and inaccuracies in Greenberg's data"; for example, Willem Adelaar, a specialist in Andean languages, has stated that "the number of erroneous forms [in Greenberg's data] probably exceeds that of the correct forms". Some forms in Greenberg's data even appear to be attributed to the wrong language. Greenberg also neglects known sound changes that languages have undergone; once these are taken into account, many of the resemblances he points out vanish. Greenberg's data also contains errors of a more systematic sort: for instance, he groups unrelated languages together based on outdated classifications or because they have similar names.
Greenberg also arbitrarily deems certain portions of a word to be affixes when affixes of the requisite phonological shape are unknown to make words cohere better with his data. Conversely, Greenberg frequently employs affixed forms in his data, failing to recognise actual morphemic boundaries; when affixes are removed, the words often no longer bear any resemblance to his "Amerind" reconstructions. Greenberg has responded to this criticism by claiming that "the method of multilateral comparison is so powerful that it will give reliable results even with the poorest data. Incorrect material should merely have a randomizing effect”. This has hardly reassured critics of the method, who are far from convincing of the method's "power".
Borrowing
A prominent criticism of mass comparison is that it cannot distinguish borrowed forms from inherited ones, unlike comparative reconstruction, which is able to do so through regular sound correspondences. Undetected borrowings within Greenberg's data support this claim; for instance, he lists "cognates" of Uwa baxita "machete", even though it is a borrowing from Spanish . Greenberg admits that "in particular and infrequent instances the question of borrowing may be doubtful" when using mass comparison, but claims that basic vocabulary is unlikely to be borrowed compared to cultural vocabulary, stating that "where a mass of resemblances is due to borrowing, they will tend to appear in cultural vocabulary and to cluster in certain semantic areas which reflect the cultural nature of the contact." Mainstream linguists accept this premise, but claim that it does not suffice for distinguishing borrowings from inherited vocabulary.
According to him, any type of linguistic item may be borrowed "on occasion", but "fundamental vocabulary is proof against mass borrowing". However, languages can and do borrow basic vocabulary. For instance, in the words of Campbell, Finnish has borrowed "from its Baltic and Germanic neighbors various terms for basic kinship and body parts, including 'mother', 'daughter', 'sister', 'tooth', 'navel', 'neck', 'thigh', and 'fur. Greenberg continues by stating that "[D]erivational, inflectional, and pronominal morphemes and morph alternations are the least subject of all to borrowing"; he does incorporate morphological and pronominal correlations when performing mass comparison, but they are peripheral and few in number compared to his lexical comparisons. Greenberg himself acknowledges the peripheral role they play in his data by saying that they are "not really necessary". Furthermore, the correlations he lists are neither exclusive to or universally found within the languages which he compares. Greenberg is correct in pointing out that borrowing of pronouns or morphology is rare, but it cannot be ruled out without recourse to a method more sophisticated than mass comparison.
Greenberg continues by claiming that "[R]ecurrent sound correspondences" do not suffice to detect borrowing, since "where loans are numerous, they often show such correspondences". However, Greenberg misrepresents the practices of mainstream comparative linguistics here; few linguists advocate using sound correspondences to the exclusion of all other kinds of evidence. This additional evidence often helps separate borrowings from inherited vocabulary; for instance, Campbell mentions how "[c]ertain sorts of patterned grammatical evidence (that which resists explanation from borrowing, accident, or typology and universals) can be important testimony, independent of the issue of sound correspondences". It may not always be possible to separate borrowed and inherited material, but any method has its limits; in the vast majority of cases, the difference can discerned.
Chance resemblances
Cross-linguistically, chance resemblances between unrelated lexical items are common, due to the large amount of lexemes present across the world's languages; for instance, English and Spanish are demonstrably unrelated, despite their similar phonological shape. This means that many of the resemblances found through mass comparison are likely to be coincidental. Greenberg worsens this issue by reconstructing a common ancestor when only a small proportion of the languages he compares actually display a match for any given lexical item, effectively allowing him to cherry-pick similar-looking lexical items from a wide array of languages. Though they are less susceptible to borrowing, pronouns and morphology also typically display a restricted subset of a language's phonemic inventory, making cross-linguistic chance resemblances more likely.