Trigger words and their effect on code-switching

Inhalt

1 Introduction: A psycholinguistic approach to code-switching

2 The triggering hypothesis
2.1 Types of triggering
2.2 The triggering hypothesis in relation to speech production models

3 Comparison of the two triggering hypotheses, their respective methodological approaches, and the evidence
3.1 Code-switching in typologically unrelated languages: Dutch-Moroccan Arabic
3.2. Codeswitching in typologically closely related languages: Dutch-English

4 False cognates as trigger words

5 Conclusion

6 References

1 Introduction: A psycholinguistic approach to code-switching

It is not unlikely for bilinguals or multilinguals to sometimes switch back and forth between two or more different languages within a single conversation or even a single sentence. This phenomenon is called “code-switching”, and it has been studied extensively from sociolinguistic and structural perspectives. In this paper, however, it will be examined from a psycholinguistic approach. As code-switching occurs rather naturally and in informal settings, psycholinguists – who usually seem to prefer working with controlled data – tend to refrain from studying the psycholinguistic aspects of this phenomenon. Thus, unfortunately, much experimental data does not seem to exist in this research field so far (Gardner-Chloros, 2009). Nonetheless, the study of code-switching with naturally occurring data could certainly provide some interesting insights into the cognitive processes of bilinguals when switching languages, and has thus been increasingly analyzed and evaluated with corpus data in recent years, e.g., by Broersma & De Bot, 2006; Broersma, 2009, whose findings will be summarized and discussed in this paper. The aim here is to present support for language switches induced by so-called “trigger words”.

The question I asked myself is “Can certain words trigger code-switching, and if so how and to what extent?” In order to discuss this, I will first introduce Michael Clyne’s (1967) triggering hypothesis, which first suggested that certain words are able to provoke a language switch in the words right next to them. Then I will relate said hypothesis to Levelt’s (1989, 1999) and De Bot’s (1992) speech production models to convey that some of Clyne’s assumptions do not quite fit with more recent knowledge about cognitive processes in speech production. After that, I will introduce Boersma and De Bot’s (2006) adjusted version of the triggering hypothesis, which takes these mental processes into account and claims that triggering occurs at the lemma level. How trigger words might influence the lemma-network of the language that is not primarily selected shall also be discussed in this section. Then I will compare and discuss the two version of the triggering hypothesis proposed by Clyne (1967) vs. Broersma and De Bot (2006) and their word level vs. clause level approaches with respect to the findings of two studies executed by Broersma and De Bot (2006) and Broersma (2009). Lastly, I will refer to a study by Broersma, Isurin, Bultena and De Bot (2009), which suggests that so-called “false cognates” can also trigger a language switch despite not sharing the same meaning.

In my conclusion, I will summarize the most important aspects of triggered code-switching based on the literature I have read and discuss whether one of the triggering hypotheses can really be seen as being superior to the other.

2 The triggering hypothesis

In his work, Michael Clyne (1967) differentiates between externally and internally induced code-switching, but the focus in this paper will lie solely on the internal factors. Clyne’s triggering hypothesis (1967, 2003) suggests that so-called “trigger words”, particularly cognates, i.e. words in different languages that are close in form and meaning, can provoke or facilitate a switch between languages within a sentence. Due to their strong similarities in said areas it is not always clear which language these items pertain to, and trigger words themselves are never classified as code-switches in the data that will later be discussed (Broersma, 2009). According to Clyne (2003), words that could act as triggers are proper nouns, lexical transfers, bilingual homophones, and compromise forms.

2.1 Types of triggering

Clyne (1967) proposes that the words most likely to undergo a language switch are the words that are closest to these cognates, i.e., the ones directly neighbouring the cognates in an utterance, as they can supposedly cause a short-term loss of the speaker’s linguistic orientation. Clyne (1967) differentiates between three triggering loci:

(a) Consequential triggering: Clyne (1967) assumes that after the occurrence of a trigger word, the speaker may briefly be confused as to which language her or she is currently engaging, and thus switch to another language with the word(s) that follow.
(1) “Ich les’ gerade eins; das handelt von einem alten SECONDHAND-DEALER and his son [...]“ (Clyne, 1987: 754)

In this example the lexical transfer “secondhand-dealer“ may act as a trigger word and thus increase activation in the non-selected language, resulting in a language switch within the utterance. The words in italics are always the code-switched words in these examples.

(b) Anticipational triggering: Here the assumption is that when speakers plan ahead what they are going to say, the production of a trigger word might subconsciously be anticipated, which could cause the speaker to switch languages right before the trigger word is articulated (Clyne, 1967).
(2) “Dann sind wir nach Warracknabeal gegangen und haben on a FARM gewohnt.” (Clyne, 1981: 37)

The word “farm may have a triggering effect here, and results in a code-switch taking place directly before the trigger word.

(c) Sandwich words: With this Clyne (1967) describes code-switched words that are in between two possible trigger words, i.e. that have trigger words both directly preceding and following them. These ‘sandwiched’ words are said to usually be conjuctions or other function words, which are mostly found between proper nouns or loanwords.
(3) “Ich war letztes Jahr in CANADA and ENGLAND”.

Here I chose to exemplify Clyne’s (1967) notion of so-called sandwich words with two proper nouns surrounding the conjunction “and” on each side. If Clyne’s (1967) assumption that a word between two trigger words has a higher chance of being code-switched is true, the conjunction could have been triggered by these proper nouns.

2.2 The triggering hypothesis in relation to speech production models

2.2.1 Levelt’s and De Bot’s speech production models

Broersma and De Bot (2006) argue that in some aspects Clyne’s triggering hypothesis (1967) contradicts current perspectives on speech production. Before explaining how their views differ, I am going to describe Levelt’s (1989, 1999) monolingual speech production model as well as De Bot’s (1992) slightly adapted model for bilingual processing, in order to provide a basic understanding of the processing components and the different levels they act on. It should be noted that when Clyne (1967) first came up with his hypothesis he lacked the knowledge on the mental processes of speech production that is available today, as he himself has acknowledged in his 2003 publication.

The first processing component in Levelt’s (1989, 1999) model is the conceptualizer, which is where the conceptual message or the idea a speaker wants to convey is generated. According to Levelt (1989), after the lexical concepts are activated, the lemma nodes which are connected to the former, get activated as well. The lemmas get selected, and a syntactic frame of a surface structure, i.e., the string of words that are to be uttered, is built. Although the order in which the lemmas will appear becomes clear at this level, the information about their form only gets available in the process of phonological encoding. Lexical, syntactic, and phonological encoding all take place at the formulation stage. This is where a speech plan containing all the information that is necessary for the articulation of the message is collected. The third stage is that of articulation. Here, the phonetic plan finally gets turned into actual speech (Levelt, 1989, 1999).

Levelt’s (1989, 1999) model only focuses on monolingual speech production. Thus, in order to consider aspects such as code-switching, De Bot (1992) extended Levelt’s (1989, 1999) model to examine the processes of bilingual comprehension and production. He proposes that within the conceptualizer, the stage of macroplanning (Levelt, 1999) – where the speaker decides on what he is going to say – is not language-specific. Based on this, De Bot (1992) argues that it is in the conceptualizer where the speaker decides on a specific language he wishes to use. The lexical information from the preverbal message then becomes lexicalized by the language-specific formulator. Furthermore, by referring to Paradis’s (1987) subset hypothesis, De Bot (1992) assumes that the lexical items of specific languages belong to different subsets which are all organized in the same language non-specific mental lexicon. According to De Bot (1992), these items can be activated to different extents.

2.2.2 Triggering effects at the surface structure level

Clyne’s hypothesis (1967) suggests that trigger words can impact a speaker’s language choice within a sentence. The trigger word’s influence here is said to be determined by where it is positioned at the surface structure of a sentence, i.e., the syntactic representation of a sentence. Clyne (1967) proposes that words right next to a trigger word may be more likely to result in a language switch.

However, the proposal that triggering takes place at the surface structure level, clashes with the current consensus on bilingual speech production models (e.g., De Bot, 1992; Grosjean, 1998; Costa & Caramazza, 1999), which postulate that language-specific lemmas are selected in one of the first stages of the speech production process, namely the conceptualization stage. Thus, language choice is said to precede the surface structure which is part of the formulation phase. Clyne’s (1967) suggestion of triggering taking place at the surface structure level implies that language choice only occurs after the conceptualization phase, i.e., in the formulation phase, which according to the speech production models mentioned above is not the case.

2.2.3 Triggering effects at the lemma level

Broersma and De Bot (2006) definitely do not negate a causal relation between trigger words and code-switching with speech production theory per se, but they adjust Clyne’s original triggering theory by relating it to the current knowledge on psycholinguistic speech production models. In Broersma and De Bot’s (2006) revised version of Clyne’s hypothesis, triggering is assumed to already occur before the formation of the surface structure. They suggest that trigger words can stimulate a change at the lemma level by boosting the activation level of the non-target language. Broersma and De Bot (2006) argue that this increase in activation could also increase the chance in the speaker’s subconscious choice of a lemma in the non-target language.

The surface structure is further devalued in terms of predicting which words might be code-switched due to a trigger word because the order of the selected lemmas does not always correlate with how the word order eventually appears in the surface structure (Broersma and De Bot, 2006).

Thus, Broersma and De Bot (2006) suggest a different method of predicting the likelihood of certain elements being code-switched. They examine two possibilities of where triggered code-switching could take place.

The first is based on bilingual speech production models where feedback from the word form to the lemma level can occur. These models (e.g., De Bot, 1993) state that trigger words are both similar in form and meaning. Their models propose that the lexical items of a specific language are part of a subset in the mental lexicon, which is not language specific (Broersma & De Bot, 2006). If word forms appear that resemble each other in two different languages they are said to be connected by a single node. If the speaker is in a setting where he or she knows that the interlocutor is also competent in the two languages he or she speaks, then it is assumed that the lemma networks of both languages are active to varying degrees. They may be at a similar level, but one language will always be at least a bit more activated than the other. Since the word form of a trigger word is shared in two languages and is connected by a node, the trigger word is said to be less active. This increase in activation can result in the less active language becoming more active than the other language – at least momentarily – and thus trigger a code-switch (Broersma and De Bot, 2006).

The other possibility or assumption Broersma and De Bot (2006) name is that feedback from word form to lemmas does not occur. This would mean that when the lemmas are selected, they do not have any information about the word forms. Consequently, there would not be any difference between trigger words and other translation pairs, i.e., words in two languages that solely share the same meaning, at the lemma level. Evidence that speaks against this theory has been found through a study by Van Hell and De Groot (1998), which shows that the conceptual representation of cognate and non-cognate translation pairs is different, and that the former can activate lemmas of the language that is not selected. According to the findings by Van Hell and De Groot (1998), there is a higher degree of conceptual overlap in cognates compared to non-cognates. Broersma and De Bot (2006) suggest that this could be the case because cognates share similarities in both meaning and form, instead of only in meaning.

Clyne (2003) would seem to argue against this interpretation by stating that the reason as to why cognates should have a higher degree of conceptual overlap than non-cognate translation pairs is uncertain, and that it is not compulsory either. Broersma and De Bot (2006) acknowledge this criticism since concreteness and grammatical class also seem to play a role in conceptual representation as found in the study of Van Hell and De Groot (1998). Therefore, they conclude that the conceptual representation between cognates and non-cognates may vary in a graded way. However, it is stated that triggered code-switching could still be explained with this assumption, as the fact that cognates and non-cognates are already treated differently before the activation of the form (Broersma & De Bot, 2006).

[...]