It was a frustrating experience reading Poplack's
Borrowing: Loanwords in the Speech Community and in the Grammar. On the one hand, it’s intelligent, well-written, and packed with a wealth of precious sociolinguistic data on borrowing and to a lesser extent code-switching; on the other hand, it appears to be largely dedicated to hammering home a definition of the former that appears to me to be fundamentally untenable. The author ably demonstrates that three criteria that one might expect to be closely correlated are not: conventionalization, morphosyntactic integration, and phonological integration are all independent of one another. Of these three, she chooses to define borrowing exclusively in terms of morphosyntactic integration. For her (enormous, but not very numerous) preferred corpora, this apparently works just fine. But...
The notion of “borrowing” emerged from diachronic studies of the vocabulary used in monolingual discourse. As such, whatever necessary criteria we choose to use to delineate marginal cases, conventionalisation must remain a sufficient criterion for borrowing: if the whole speech community uses the form irrespective of individuals’ level of competence in its source language, it must be a borrowing, not a code-switch. Poplack rejects the criterion of conventionalization as essentially extra-linguistic, preferring the criterion of morphosyntactic integration; yet the latter invokes community conventions just as much as the former, the only difference being the type of conventions invoked (grammatical vs. lexical.) Finding that single words of foreign origin overwhelmingly display morphosyntactic integration and are thus by her definition nonce borrowings, she concludes (p. 213) that “loanwords do not originate as code-switches… the very first mention of a nonce form already features the full complement of morphosyntactic integration into [the recipient language]”. But this makes some problematic predictions.
First of all, if this is true, borrowings should never retain source morphosyntax. This is clearly not tenable. Borrowings retain source morphology all the time: Berber nouns in several Arabic dialects, and Arabic nouns throughout Berber, keep their plurals; Latin nouns in German keep their case markers; in a tiny scattering of languages around the Mediterranean, such as Ghomara Berber, borrowed verbs even keep their conjugation. Some categories of borrowings retain their syntax as well: larger borrowed numerals precede or follow the noun according to the rules of the source language, not of the recipient, in Korandje; borrowed primary adpositions and complementizers rather consistently place their complement as in the source language wherever they are found (cf. Moravcsik 1978). Poplack attempts to dispose of the latter with a short footnote (p. 50): “More wide-ranging proposals for borrowability hierarchies […] including prepositions, determiners, pronouns, clitics, and complementizers may be characteristics of certain extreme borrowing situations, such as pidginization or creolization, or, alternatively, the result of confounding code-switches […] and borrowing. The latter is so heavily restricted to content words that this is practically a defining characteristic.” But this really will not do. Turkish (which has borrowed the complementizer ki from Persian along with the associated word order) is hardly anyone’s idea of a pidgin or creole!
Second, such a claim (along with the book as a whole) seems to presuppose that borrowings are necessarily single lexical items. This is manifestly not the case. In English, borrowings that consist of multiple source language words (quid pro quo, per cent, hors d’oeuvres…) are sufficiently unanalysable to be considered as single lexical items in the recipient language; these need not pose a problem for Poplack. But in quite a few languages, including many Berber varieties, at least two classes of multi-word borrowings remain clearly analysable as multiple words, and productive, even for monolingual speakers: numerals, and numeral+measure noun combinations. Such borrowings must necessarily start out as code-switches in Poplack’s terms.
From these facts, I conclude that the process of conventionalization is even more independent of morphosyntactic integration than Poplack assumes. Morphosyntactic integration, as Myers-Scotton implies, is far stricter for structure than for semantics, and is strictly obligatory in neither case. And for function words, at least, syntactic integration only concerns relations up the tree, not down it. It follows that neither morphosyntactic nor phonological integration can be considered necessary or sufficient criteria for borrowing.