Somebody at a conference in Leeds told a room of us that she code-switches when she rings her mother, and everyone nodded, and what she meant was that she softens her vowels. That is a real thing which really happens to her, and it has a name, and the name is not code-switching.
I would like the term back. Not out of pedantry, or not only. The technical sense is a better story than the metaphor, and the metaphor has been getting the applause.
The definition goes back at least to Weinreich in 1953, who described bilinguals as people who switch from one language to the other according to appropriate changes in the speech situation. Muysken, in a 2000 typology, separates two things: codeswitching as "the rapid succession of several languages in a single speech event", and code-mixing as "all cases where lexical items and grammatical features from two languages appear in one sentence". Usage is genuinely inconsistent in the literature, which is worth knowing before anyone gets confident: some writers reserve code-mixing for switches inside a clause, some do the reverse, some treat the pair as synonyms.1 The one distinction almost everybody keeps is between switching between sentences and switching inside one. Intersentential and intrasentential. The second is where the interesting problem lives.
Here are three from my own family, glossed. Arabic in Levantine transliteration.
1 Don't, ʿēb.
ʿēb = shame, disgrace (interjection)
A single item at a clause boundary. Nothing in the English
grammar has to accommodate anything.
2 He is completely tʾīl id-damm and everyone knows it.
tʾīl id-damm = heavy-of the-blood, i.e. tedious
The Arabic arrives whole, carrying its own definite article
and its own head-first order. English supplies the copula,
the adverb, the tense and the conjunction.
3 yiʿṭīk el-ʿāfye, love, I'll be two minutes.
yiʿṭīk el-ʿāfye = may-he-give-you the-health
A complete Arabic formula, then the rest of the turn in
English. Two grammars, neither of them damaged.
Number two is the one to look at. Myers-Scotton's Matrix Language Frame model would call that Arabic phrase an embedded language island: a constituent that comes in intact, internally run by its own grammar, sitting inside a clause whose morpheme order and grammatical morphemes are supplied by the other language.2 The English frame is not confused about what it is doing. It has taken a foreign object and given it a job.
Which brings us to the constraints, and to the fact that they leak.
Poplack's 1980 paper in Linguistics is the founding document here, and it proposed that
code-switches will tend to occur at points in discourse where juxtaposition of L1 and L2 elements does not violate a syntactic rule of either language
with the follow-on that a switch is blocked inside a constituent built by a rule only one of the two languages has. The free morpheme constraint came alongside it, and the sharpest wording is Sankoff and Poplack's the following year, forbidding a switch
between a bound morpheme and a lexical form unless the latter has been phonologically integrated into the language of the bound morpheme
Both have been broken in print, repeatedly, and Arabic keeps doing the breaking. Rabia Redouane tested both of these on Moroccan Arabic and French speakers in Canada, along with a third constraint about the size of the switched constituent, and found switches between verb and object and between noun and adjective where the two surface structures do not match at all, plus switches across word-internal morpheme boundaries, an Arabic inflection sitting on a French stem. She is scrupulous that she had four speakers, and four speakers is four speakers. But Bentahila and Davies had already reported French verb stems wearing Arabic affixes in Moroccan corpora, and Berk-Seligson found a Hebrew stem, talfén, to phone, taking a Spanish verb ending.
Now look at what a violation is made of. To put an Arabic inflection on a French stem you have to know the Arabic verb pattern well enough to apply it to a root that pattern has never met, pick the right template, and get the internal vowels right, in the middle of a sentence you are also steering in French. Nobody does that by accident. The examples that break the theory are harder than the examples that confirm it. That is the finding I want people to take away, and it is the opposite of what the folk theory says: fluent switching is not a failure to keep two languages apart, it is a demonstration that both grammars are online and being satisfied simultaneously at the seam.
There is a bonus in Poplack's condition, which is that switching wants the surface structures to map onto each other. Translation is the trade conducted entirely in the places where they do not. @marginalia/catullus-85-in-eleven-words is a few words of Latin that will not sit down in English, and the reason is precisely the reason a switch would be blocked there: no joint. A bilingual finds the joints. A translator has to build one, out of materials that do not fit, and then live with it.
So when the woman in Leeds said code-switching, she was describing shifting within one language, which the variationists have been mapping since the 1970s, Labov's name on most of it, Trudgill's on a good chunk of the rest. Interesting work. Different object. The metaphor got adopted because it sounds technical, and it has quietly cost us the technical thing, which was better: your grandmother, who left school at twelve, is running two grammars at once and satisfying both of them at the joint, at conversational speed, and has never once been told this is difficult.
Nobody puts that on a slide.
The safest habit is to name the phenomenon rather than the label: say intrasentential switch, or insertion, or say a switch at a clause boundary, and nobody has to guess which school you belong to. ↩
Myers-Scotton, Duelling Languages: Grammatical Structure in Codeswitching (Clarendon, 1993). The model is asymmetric by design: one participating language, the matrix, supplies the morphosyntactic frame of the bilingual clause, and the other, the embedded language, contributes into it. Two principles do most of the work, and I am paraphrasing rather than quoting: in a mixed constituent, surface morpheme order follows the matrix language, and the grammatical morphemes that hold the clause together also come from the matrix language, except inside an embedded language island, which is exempt because it is internally governed by its own grammar. The model has serious competition, including Muysken's typology of insertion, alternation and congruent lexicalization, which does not assume a matrix language at all. Do not treat any of this as settled. ↩