What decides the language family of a language the most structure/grammar or the vocabulary?

Question

My assumed premise:

Indo-European language classification is broad. We can always find two languages of this family which are grammatically so different, and also the languages grammatically similar.

But most of the study of this language family is based on comparative vocabulary and its cognates.

But

Most of the south Indian Dravidian languages have many Sanskrit words in them from times immemorial(it's not clear). For example Telugu vocabulary is mostly Sanskrit words that a Telugu person cannot(might not) give you a pure Telugu equivalent(Dravidian equivalent? of most of the words. It also attained morphological features and grammar features like Sandhi. Strangely enough now the trend is English words.

But it still is not classified under(not even in a partial sense) Sanskrit. The reason given is the language is not its words but its grammar. But the Indo European languages are classified mostly due to similarity in their words.

I understand this may be due to the fact that the Sanskrit words in Telugu, no matter when they might have been added, are loan words.

In this respect, I'm wondering what's the clear line between using grammar vs words in classifying world languages. Is it consistent across all languages? Or some language families have different peculiarities for their languages that made them under same language family?

For example for Dravidian languages, is it their grammar?

And also could it be possible that a language got influenced by other languages so much in many respects that it got classified in different language family than it should be? (Of course this might be ambiguous)

Answer 1

Your question isn't entirely clear, and Greg Lee has implicitly answered one version, namely how do we determine the subgrouping of languages that we know to be related, for example how do we know that Hindi and Farsi are more closely related that Hindi and English – shared grammatical innovations. Typically there are more innovations in the form of phonological rules, but subgrouping can be based on morphological or syntactic innovations.

But this starts from the assumption that the languages in question are related: then the question is, how do we know whether two language are in fact related. As you point out with the case of Telugu, there is a lot of similarity (of a type) between Telugu and Sanskrit, so one could hypothesize that the languages descend from some common ancestor. This is where the role of grammar becomes especially important. There is actually very little grammatical similarity between Telugu and Sanskrit. If you compare Sanskrit, Avestan, Ancient Greek, Latin, Gothic and so on, you find massive similarity in the actual case suffixes and verb inflections, to the point that we can reconstruct grammatical morphemes for the proto-language and we can see that regular sound changes apply in the development of these case suffixes. Comparing the case suffixes of Sanskrit and Telugu, well, Sanskrit has them, Telugu doesn't, and the Telugu postpositions don't at all resemble the Sanskrit case suffixes. (The existence of sandhi is not particularly probative since many languages have rules of sentence phonology; there is little similarity between the languages in the actual rules of sandhi -- although, the Sanskrit sandhi system has been taken wholesale as part of compounding, so this is actually an area where aspects of grammar can be borrowed).

The underlying premise, which has been a standard assumption in linguistics for centuries, is that it is easy to borrow words, and hard (though not impossible) to borrow grammar. English now has a bunch of words borrowed from Hindi and various Dravidian languages, and has not borrowed a single item of grammar. One of the most important early works on reconstruction of Indo-European by Franz Bopp was his 1816 On the Conjugation System of Sanskrit in comparison with that of Greek, Latin, Persian and Germanic, which preceded his work on sound laws.

It is true that historically speaking, the idea of language-relatedness is often based on lexical similarity and the ancient idea of there being a connection between Greek and Latin, Arabic and Hebrew, Saami and Hungarian was based on word similarity. Nevertheless, it is well-recognized that similarity on lexicon can be quasi-accidental (specifically, due to borrowing of words rather than genetic relationship), and it is a standard methodological principle that only via a comparison of aspects of grammar, rather that just lexicon, can genetic relatedness be established.

Answer 2

The basis of the traditional classification of IE, since the 19th century, has been neither shared grammar nor shared vocabulary, but rather shared sound changes.