Greenberg's reputation rests in part on his contributions to synchronic linguistics and the quest to identify linguistic universals. In the late 1950s, Greenberg began to examine corpora of languages covering a wide geographic and genetic distribution. He located a number of interesting potential universals as well as many strong cross-linguistic tendencies.
In particular, Greenberg invented the notion of "implicational universal", which takes the form, "if a language has structure X, then it must also have structure Y." For example, X might be "mid front rounded vowels" and Y "high front rounded vowels" (for terminology see phonetics). This kind of research was taken up by many scholars following Greenberg's example and continues to be an important kind of data-gathering in synchronic linguistics.
Like Noam Chomsky, Greenberg sought to discover the universal structures underlying human language. Unlike Chomsky, Greenberg’s approach was empirical rather than logico-deductive. Greenberg’s approach, often characterized as "functionalist", is commonly opposed to Chomsky’s rationalist approach. A call to reconcile the Greenbergian and Chomskyan approaches can be found in Linguistic Universals, edited by Ricardo Mairal and Juana Gil (2006). It remains to be seen whether this call will be heeded.
Many who are strongly opposed to Greenberg's methods of language classification (see below) nevertheless acknowledge the importance of his typological work, in particular his tremendously influential 1963 article, "Some universals of grammar with particular reference to the order of meaningful elements."
Greenberg proposed a controversial method for finding historical relationships when comparing too many languages for traditional methods of establishing regular sound shifts to be practical – a situation that arises particularly when attempting to establish long-range historical families in regions of the world where few if any lower-level families have been reconstructed, or where linguistic diversity is especially high. This method was enthusiastically embraced by a few historical linguists (and many geneticists), though rejected as pseudoscience by most historical linguists. See mass lexical comparison and Amerind for a fuller discussion.
Using the method of mass comparison, Greenberg arrived at a number of novel classifications of languages. All these classifications were rejected when first proposed as factually incorrect and methodologically unsound. Since then, some have come to be accepted in whole, others in part; some are still rejected; the status of others is pending. Some details and appreciations of these classifications follow.
Greenberg is widely known for his development of a new classification system for African languages, which he published in 1963. The classification was for a time considered very bold and speculative, especially in his proposal of a Nilo-Saharan language family, but is now generally accepted among African historical specialists. In the course of this work, Greenberg coined the term "Afroasiatic" to replace the earlier term "Hamito-Semitic" after showing that Hamitic is not a valid language family.
Greenberg's classification was largely based on earlier classifications, making new macrogroups by joining already established families - based on his method of mass comparison. The classification has been used as a basis for further work and some historical linguists have proposed even broader proposals of African language families. Hal Fleming introduced the Omotic family, and Gregersen proposed the joining of Niger-Congo and Nilo-Saharan into a larger Kongo-Saharan family, which were in turn accepted by Greenberg, though in the case of Kongo-Saharan only implicitly.
Greenberg's work on African languages has been criticised by historical linguists Lyle Campbell and Donald Ringe, who do not feel that his classification is justified by his data and request a reexamination of his macro-phyla by "reliable methods" (Ringe 1993:104). Even Fleming and Lionel Bender, who are sympathetic to Greenberg's classification, acknowledge that at least some of his macrofamilies (particlularly Nilo-Saharan and Khoisan) are not fully accepted by the linguistic community and may need to be split up (Campbell 1997).
In 1971 Greenberg proposed the Indo-Pacific superfamily, which groups together the Papuan languages (several language families spoken in Papua New Guinea and nearby regions which are not Austronesian) together with the native languages of Tasmania and the Andaman Islands, but excludes Australian Aboriginal languages. This grouping is considered by most linguists to be highly speculative and is not accepted by anyone working on Papuan or Tasmanian languages.
Americanist linguists classify the native languages of the Americas into two large families, Eskimo-Aleut and Na-Dené, spoken in well-defined parts of North America, and some 600 to 2000 other separate families (Diamond 1997:368), spoken in the rest of North America and through Central and South America. Early on, Greenberg (1957:41) became convinced that many of the reportedly unrelated languages could be classified into larger groupings. In his 1987 book Language in the Americas, he supported the Eskimo-Aleut and Na-Dené groupings, but proposed that all the other Native American languages belonged to a single family. He termed this postulated family Amerind.
This theory is rejected and has been soundly criticized by most historical linguists. The criticisms are directed not so much toward the classification per se, but primarily to the method of mass lexical comparison used to establish it, which the majority of historical linguists consider inherently unreliable (see above); and toward the large number of errors that have been shown to be present in the sources used by Greenberg, such as wrong or non-existent words, incorrect translations, words attributed to the wrong languages, and unsupported or wrong identification of prefixes and suffixes.
While some of these errors (which, according to Greenberg's defenders, only affect a few percent of the data) could conceivably lead to an artificial increase in the similarity measure, others would merely introduce random noise in the measurement, and therefore tend to reduce it — which would only strengthen Greenberg's conclusions. Nevertheless, the allegations of widespread errors in the data along with objections to his methodology have led most linguists to dismiss this part of Greenberg's work as unscholarly and invalid.
Later in his life, Greenberg proposed to join many language families of Europe and Asia into a single group called Eurasiatic, fairly similar to Illich-Svitych's earlier Nostratic proposal but differing in important ways - notably the exclusion of the Afro-Asiatic languages, which has since become popular among Nostraticists as well. He continued to work on this project from the time of his diagnosis with fatal pancreatic cancer until his death.