From Greenberg (1957: 42–43):
All available grammatical information should be systematically examined, but vocabulary leads most swiftly to the correct hypotheses as a general rule. The effectiveness of mass comparison of basic vocabulary, for all its apparent simplicity, is illustrated in Table 3 by only a few forms from all the contemporary languages of Europe. Note that, even by the time the second word has been examined, the correct hypothesis emerges.
Not impressed by this assertion. Sure the comparison of just '1' and '2' will easily suggest some language groups like Germanic, Baltic, Slavic, Finnic, and leave Basque so far isolated. But Hungarian egy (also mistranscribed here as /ed/ and not /ɛɟ/ or even /edʲ/) will not be readily identified as (and probably isn't) cognate to Finnic *üksi. One might rather end up thinking it is a reduced reflex of Slavic #jadin (actually *edinъ but we will not exactly be doing any real reconstruction based on just this data). Két also shows no highly compelling similarity to *kaksi; maybe would do so more if G. had bothered to inform that the inflected stem is kahte-. So just by comparing the first two numerals' citation forms, the most sound conclusion might be that Hungarian is either a divergent Slavic language, or an isolate that has borrowed '1' from Slavic!
Do we really retrieve correct sorting of the other IE languages from just these either, if we pay no heed to G. having already sorted the data according to what we should think? I see no consistent way to divide Celtic from Romance, rather e.g. Irish and French seem to come close to each other. And nothing whatsoever even suggests Balto-Slavic. Instead, I think, we'd end up with a basic division Slavic / Rest by different words for '1', which then suggests that the cluster *tv- in Germanic is archaic and that groups with only *d- are a single unit; and Greek, of course, subgroups with Baltic by virtue of suffixal -as in '1' and evidence for a close front element in '2'. Maybe it would remain there even after considering the rest of thsi data, thanks to e.g. #afci 'ear', #(m-)aki 'eye', and the i-suffix in #dand-i 'tooth'!
This has probably been noted before, just adding to the record.
