Presentation on theme: "1 Code-switching in a network Richard Hudson Bangor February 2008."— Presentation transcript:
1 Code-switching in a network Richard Hudson Bangor February 2008
2 Linguists and psycholinguists linguists psycho-linguists Different worlds, no shared models We need a bridge.
3 Lingualism Monolingualism –even monolinguals know about other languages Bilingualism Multilingualism –World record for societal multilingualism = Hyperpolyglottism –World record for individual multilingualism = ? ? 6 72?
4 Theory How do monolinguals record knowledge about languages? –e.g. Latin is dead How do linguals keep their languages separate? –e.g. Dog is English, ci is Welsh How do these facts relate to language- structure?
5 Languages in the mind Each named language is a concept. Its part of general knowledge. General knowledge is a network of concepts. So each named language is a node. And its properties are links to other nodes.
6 Latin is dead Latin language spoken by community 0 isa exception default
7 Word Grammar Language competence is a network too. –not a network of lexical items or constructions –nodes have no internal structure. Its just our (ordinary) knowledge of words. So (1) it has no boundary. So (2) activation spreads freely between language and non-language.
8 Spreading activation Language must be a network because it carries spreading activation. Evidence: –Priming: word 1 primes word 2 if they are network neighbours. –Speech errors: the substituted word is activated by accident from the target or context.
13 Priming at all levels Words prime network neighbours in: Phonology: verse primes nurse (but only briefly) Morphology: hedges primes hedge for longer than pledge does. Syntax: Vlad brought a book to Boris primes other V + DO + PP sentences Semantics: nurse primes doctor.
14 How activation spreads Blindly – hence errors. In any direction, depending on the target. –Speaking or listening or analysing or … Randomly, so weak activation has a weak effect (rather than no effect) –because activation converges from many directions.
15 No boundaries Activation flows freely between language and non-language Even speech errors may have non-linguistic causes –E.g. (By a computer) Do you have a Choice of language is another example. computer ? screwdriver ?
16 Languages in a network Each word has (belongs to) a language. Each language has a prototypical word. Prototypical words are the domain of phonology.
17 Welsh and English Welsh English language Welsh-word English-word language ci diolchdog thanks
18 A language isnt a box. WelshEnglish cidiolchdogthanks
19 Pro network, con boxes Translation equivalents in different languages prime one another. –e.g. ci primes dog. So they must share meaning. Likewise for phonological or graphological priming. –e.g. ci primes key or cipher.
20 Cross-language links in a network ci dog Welsh-word English-word noun key meaning /ki:/ sound
21 Moreover, L1 and L2 are unequal Often, L2 was learned via L1 –e.g. Ci means dog So some L2 words are linked directly to L1 words by a translation relation, but not vice versa. This explains why L2 > L1 translation is faster and easier than L1 > L2 (Kroll & Dussias 2004). But in time L2 becomes independent of L1.
22 The asymmetry of L1 and L2 ci dog L2-word L1-word translation meaning
23 Language choice Activation from meaning activates words in both languages. So how does a bilingual stick to one language when in monolingual mode? Two suggestions so far (Costa 2004): Non-target language activation is: –inhibited. –ignored. Both assume an external controller.
24 Code-switching with external control ci dog Welsh-word English-word meaning /ki:/ sound /d g/ sound winner? suppress or ignore
25 External control? Who is the external controller? –a homunculus? –problem: infinite regress. What does he control? –just language use? If possible, avoid special external controls.
26 Code-switching with internal control ci dog Welsh-word English-word meaning /ki:/ sound /d g/ sound winner?
27 Two kinds of code-switching Situational or intra-sentential. Bilingual speakers can stick to one language when the situation requires it –e.g. when speaking to a monolingual How do they do it? The situation keeps the required language node active.
28 Speaking to a Welsh speaker Welsh English language Welsh-word English-word language-of ci dog spoken by Welsh speakers current interaction addressee
29 Intra-sentential code-mixing Used only when speaking to bilinguals. Bilinguals belong to both communities. So the situation activates both languages. e.g. (Eppler 2004) –und heuer fahren wir nach Harrogate for a long-es weekend –die do-'nt mind aber I do.
30 Speaking to a Welsh-English bilingual Welsh English Welsh-word English-word language-of ci dog spoken by Welsh speakers current interaction addressee English speakers spoken by
31 Random choice Suppose each language is equally active. So for each word each language is equally likely: –like tossing a coin, where each toss is independent of the previous one: –A A B A B B A A A B B A B B But thats not how code-mixing works.
32 Inertia prevails Words tend strongly to be in the same language if they are: –adjacent or –linked by a syntactic dependency. Why does adjacency matter? –the previous language is still most active Why does dependency matter? –the dependency link carries activation.
33 Why does the language tend to stay the same? Lang A Lang B word 1 word 2 word 3 language dependent or head a concept meaning ? ?
34 Conclusion Language choice is governed by activation of: –a required language node to match the addressees social category –the previous word –a syntactically related word Language is integrated into the network of general knowledge.
35 The bridge is growing linguistics psycho-linguistics Computer model with numbers
36 Diolch yn fawr For this slide show: www.phon.ucl.ac.uk/home/dick/talks.htm#bangor For more about hyperpolyglots etc: …dick/polyglotism/home.htm For more about Word Grammar: …dick/wg.htm