Origin of the Nostratic theory
Pedersen seems to have first used the term "Nostratic" in an article on Turkish phonology published in 1903. The kernel of Pedersen's argument for Nostratic in that article was as follows (1903:560-561; "Indo-Germanic" = Indo-European):
Grønbech considers it possible p. 69 that the Turkish word for "goose" could be borrowed from Indo-Germanic (Osm. kaz Yak. xās Chuv. xur). There are in my view three possibilities with regard to this word: coincidence, borrowing, and kinship. One must also reckon with this last possibility. Very many language stocks in Asia are without doubt related to the Indo-Germanic one; this is perhaps valid for all those languages which have been characterized as Ural-Altaic. I would like to unite all the language stocks related to Indo-Germanic under the name "Nostratic languages." The Nostratic languages occupy not only a very large area in Europe and Asia but also extend to within Africa; for the Semitic-Hamitic languages are in my view without doubt Nostratic. With regard to the proof of the relationship of the Nostratic languages, not only must all root etymologies and in general all etymological frivolities be kept at a distance, but one should in general not concern oneself with heaping up a mass of material. One should rather limit oneself to the rational consideration of a series of pronouns, negatives, in part also numerals which can be traced through several language stocks (in Turkish one is reminded of the Indo-Germanic by the negation -ma, -mä and the word-initial interrogative particle m, the interrogative pronoun kim, the pronoun of the first person män, the verbal ending of the 1. sing. -m, 1. plur. -myz, -miz and the ending -jin in the 1. sing. of the "optative," very reminiscent of the Indo-Germanic subjunctive [with the optative affix -a-, -ä-], the pronoun of the 2. sing. sän [cp. the IdG. verbal ending -s], the causative formation with -tur- [cp. IdG. -tōr nomen agentis; the Indo-Germanic causative also appears as if it were derived from a nomina agentis of the φορός type], the nomina actionis like Orkh. käd-im "clothing," several numerals: Orkh. jiti "7," jitm-iš "70," [with j = IdG. s as in Proto-Turk. *jib- "approach," Osm. jyldyz "Star": to Indo-Germanic word for "sun," jat- "lie": IdG. word for "sit"]; Proto-Turk. bǟš "5" [with š = IdG. -que; cp. Osm. piš- "be cooked," IdG. *pequeti "cooks"] etc., etc.). I resist the temptation to enter into this question in more detail.
Pedersen’s last sentence should be understood as referring to the article he was writing, not the rest of his career. Although he defined the Nostratic family, he himself never produced the work of synthesis the concept seemed to call for. That would await the work of the Russian scholars Illich-Svitych and Dolgopolsky in the 1960s for its first iteration. Nevertheless Pedersen did not abandon the subject. He produced a substantial (if overlooked) article on Indo-European and Semitic in 1908. He produced a detailed argument in favor of the kinship of Indo-European and Uralic in 1933. In effect, the three pillars of the Nostratic hypothesis are Indo-Uralic, Ural-Altaic, and Indo-Semitic. Pedersen produced works on two of these three, so the impression is incorrect that he neglected this subject in his subsequent career. His interest in the Nostratic idea remained constant amid his many other activities as a linguist.
English "Nostratic" is the normal equivalent of German nostratisch, the form used by Pedersen in 1903, and Danish nostratisk (compare French nostratique). His 1931 American translator rendered nostratisk by "Nostratian," but this form did not catch on.
In his 1931 book, Pedersen defined Nostratic as follows (1931:33
:
As a comprehensive designation for the families of languages which are related to Indo-European, we may employ the expression Nostratian languages (from Latin nostrās "our countryman").
In his view, Indo-European was most clearly related to Finno-Ugric and Samoyed, with "similar, though fainter, resemblances" to Turkish, Mongolian, and Manchu; to Yukaghir; and to Eskimo (1931:33
. He also considered Indo-European might be related to Semitic and that, if so, it must be related to Hamitic and possibly to Basque (ib.).
In modern terms, we would say he was positing genetic relationship between Indo-European and the Uralic, Altaic, Yukaghir, Eskimo, and Afro-Asiatic language families. (The existence of the Altaic family is controversial, and few would now assign Basque to Afro-Asiatic.)
However, in Pedersen's view the languages listed did not exhaust the possibilities for Nostratic (ib.):
The boundaries for the Nostratian world of languages cannot yet be determined, but the area is enormous, and includes such widely divergent races that one becomes almost dizzy at the thought. (...) The question remains simply whether sufficient material can be collected to give this inclusion flesh and blood and a good clear outline.