Wisdom? I say old, you say tooth
On the conceptual transfer of the multilingual mind:
associative priming in English
Honours dissertation
Master in English Language and Culture Faculty of Arts
University of Groningen
Bregtje J. Seton S1332236
Supervisor: Dr. W.M. Lowie Second reader: Dr. M.H. Verspoor Date: 31 August 2010
Word count: 15,252
AA
AA C K N O W L E D G E M E N T SC K N O W L E D G E M E N T SC K N O W L E D G E M E N T SC K N O W L E D G E M E N T S
It goes without saying that this thesis could not have been written without the help and support of certain important people. First of all, I would like to thank my supervisor, Dr. Wander Lowie, for his infinite patience and support. He was always willing to attend to the problems I encountered and he would always help me try to answer the questions I could not answer myself. A big note of thanks also goes to the second reader of this thesis, Dr. Marjolijn Verspoor, for her words of encouragement, her helpful comments, and her ever so innovating ideas. I must also add that most of what I have learned and (more importantly) remembered about linguistics in the past eight years has mostly been the work of both Marjolijn and Wander. I am extremely grateful for all they have taught me over the years, and for all the opportunities they have given me. Apart from having supervised my thesis, they have been indispensably important in my career as a student of English and Linguistics.
The various teachers at the English department receive my thanks for allowing me to use their valuable teaching time to ask students to participate in the experiment. I am also grateful to the employees of all the international offices of the University of Groningen for forwarding my e-mails again and again to the native speakers of English in their faculties. I’d also like to thank members of the Experimental Linguistics Group for listening to my talk and spending time to give me some of their valuable advice. I thank Prof. Dr. Huub van den Bergh and Dr. Gerben Mulder for discussing my data at the lot course on statistics and reliability. Finally, Margreet van Koert and Connie Lahmann deserve a big thank-you for their helpful comments. And of course, many thanks go to all of my participants, who were so nice to lend me some of their time to participate in my experiment.
I would like to thank my parents for supporting me unconditionally, despite not really knowing what it was they were supporting. I should thank my little brother for being the wonderful person he is, and for always making time for me whenever I need him. Last but not least, I would like to thank my friends, who always listened when I needed to talk, who tried to motivate me when I needed to work, and who made me take time off when I needed to relax. Connie, Irene, Josbert, Joyce, Luuk, Margriet, Marjella, Mirthe, Natalia, Petra, Renske, Sieuwke, Sybrine, Sylvain, Wolter, and all my other friends: thank you!
L I S T O F A B B R E V I A T I O N L I S T O F A B B R E V I A T I O N L I S T O F A B B R E V I A T I O N L I S T O F A B B R E V I A T I O N SSSS
bnc British National Corpus dst Dynamic Systems Theory
eat Edinburgh Word Association Thesaurus en prime Prime based on English-like association erp Event Related Potential
icm Idealised Cognitive Model
l1 First Language
l2 Second Language
ldt Lexical Decision Task
LSA Latent Semantic Analysis
nl prime Prime based on Dutch-like association rhm Revised Hierarchical Model
rt Reaction time
sdt Signal Detection Theory soa Stimulus Onset Asynchrony
S U M M A R Y S U M M A R Y S U M M A R Y S U M M A R Y
Word association research has found qualitative differences between native speakers and second language speakers, in that second language speakers map conceptual meanings of L1 words onto the L2 words. Long stimulus onset asynchronies (soas) in associative priming studies have found priming differences between native speakers and second language speakers. The current study was interested in whether conceptual transfer would take place for Dutch students of English during 4 years of studying English at a Dutch university. Another goal was to look at the influence of short soas in an associative priming study. Word associations from native speakers of Dutch and from native speakers of English were used to form two counteracting priming conditions for an associative priming study with a lexical decision task. Participants were a low proficiency group of Dutch students of English, a high proficiency group of Dutch students of English and a group of native speakers of English. Results showed great variation between different (groups of) items, suggesting that highly proficient students of English perform native-like on items which have different connotations in Dutch and English, and that native speakers of English show inhibitory responses on items with backward associative strength, whereas the Dutch students did not show inhibition on these items. A short soa does not show as much priming as a longer soa, which may mean that conceptual information has not been fully accessed yet.
T A B L E O F C O N T E N T S T A B L E O F C O N T E N T ST A B L E O F C O N T E N T S T A B L E O F C O N T E N T S
ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS
ACKNOWLEDGEMENTS...1
LIST OF ABBREVIATION LIST OF ABBREVIATION LIST OF ABBREVIATION LIST OF ABBREVIATIONSSSS...2
SUMMARY SUMMARY SUMMARY SUMMARY...3
TABLE OF CONTENTS TABLE OF CONTENTS TABLE OF CONTENTS TABLE OF CONTENTS... 4
1. 1. 1. 1. INTRODUCTION INTRODUCTION INTRODUCTION INTRODUCTION...5
2.2. 2.2. THEORETICAL BACKGRO THEORETICAL BACKGROUND THEORETICAL BACKGRO THEORETICAL BACKGROUNDUNDUND... 8
2.1 What are concepts? ... 8
2.2 How are concepts formed? ... 9
2.3 How do concepts differ? ...11
2.4 What happens to the conceptual representation of bilinguals? ... 12
2.5 Word association studies...16
2.6 Semantic or associative priming and other evidence... 17
2.7 Summary and present study... 21
3. 3. 3. 3. METHOD METHOD METHOD METHOD... 23
3.1 Word associations... 23
3.1.1. Participants... 23
3.1.2. Materials... 23
3.1.3. Procedure...24
3.1.4. Design and Analyses...24
3.2 Associative Priming Experiment...26
3.2.3. Participants...26
3.2.2. Materials...28
3.2.3. Procedure...29
3.2.4. Additional variables...30
3.2.5. Analyses... 31
3.2.6. Hypotheses... 32
4.4. 4.4. RESULTS A RESULTS A RESULTS A RESULTS AND ANALYSESND ANALYSESND ANALYSESND ANALYSES...34
4.1 Correlation of vocabulary scores and EN prime RTs ...34
4.2 Overall group differences...34
4.3 Analysis of frequency... 35
4.4 Analysis of prime-target relationship...36
4.5 Extra variables... 37
4.6 Qualitative Item analysis... 37
4.6.1 Items with different connotations in English and Dutch...38
4.6.2 Items with a backward relationship for only the EN prime and target...39
4.6.3 General items with no differing connotations in English and Dutch...41
4.6.4 Items that should not have been used...41
5.5. 5.5. DISCUSSION DISCUSSION DISCUSSION DISCUSSION...43
6.6. 6.6. CONCLUSION CONCLUSION CONCLUSION CONCLUSION...47
6.1 Suggestions for further research... 48
REFERENCES REFERENCES REFERENCES REFERENCES... 50
APPENDICES APPENDICES APPENDICES APPENDICES... 55
Items with different connotations in English and Dutch... 55
Items with a backward relationship for only the EN prime and target...56
General items with no differing connotations in English and Dutch...59
Not very good prime-target pairs... 64
1 .1 .
1 .1 . I N T R O D U C T I O N I N T R O D U C T I O N I N T R O D U C T I O N I N T R O D U C T I O N
In conversations with other people, negotiation of meaning takes place by relying on common ground. This common ground is mostly created by similarities in our conceptual systems. That is why a conversation with a neighbour whom we have known our whole life is less likely to give cause to misunderstandings than a conversation with someone whom we have just met. Such misunderstandings are even more likely to occur when speaking to someone in a language that is not our first.
Even though one might speak a language rather fluently, lexical items1 in that language could have certain connotations that one is not aware of. For example, a Dutch learner of English may not know that gravy often refers to a thick sauce in English-speaking countries, because the Dutch counterpart of the word (jus) is usually much more watery. Thus Dutch people will probably have a different association with gravy than native speakers of English do. Words are in fact only labels for the underlying representations, but these underlying representations may be different from one another. The more distance there is between two individuals in terms of culture and language, the more their underlying conceptual systems will differ from one another.
The closest we can get to measuring these differences is by trying to put concepts into words. Since every conceptual representation – be it an image, a thought, a feeling, a memory, a smell, a movement – is in principle an association with a word, they could be verbalized by taking word associations: the first word that comes to mind.
The present study starts at the assumption that word associations can be a way of gaining insight into conceptual representations. In language research, word associations have shown to be different for speakers of different languages (e.g.
Verspoor, 2008), even if it concerns direct translations of the same lexical item. This result maps on to the idea that speakers of different languages have different concepts of the same words. Bilinguals or second language speakers may then have concepts that are blended together from the languages or cultures they belong to.
Although it is acknowledged that different languages have words with slightly different meanings or that one language has one word for which another language has
1 It should be mentioned that the notions of a word and a lexical item (or lexeme) are often used interchangeably here. However, one should be aware that a word can always be a lexical item but that a lexical item can exist of multiple words conveying one meaning, as for example the lexical item kick the bucket, which as a string of words is a synonym of the verb to die.
multiple words, the amount of influence of a language and its different conceptual representations on a speaker of another language has not been studied to a great extent.
There are two theoretical perspectives that have been taken by different sides of bilingualism research: the language selective access approach that assumes that there are separate lexicons (e.g. Gerard & Scarborough, 1989) and the language non-selective access approach assuming that lexical items are connected into something that we could call a multilingual mental lexicon (e.g. Dijkstra, 2005). Since much cross- language priming has been found in the past few decades (e.g. Lemhöfer et al., 2008), the present study will take the latter approach. A logical consequence from taking this perspective is that conceptual information of different languages should also be linked through this multilingual lexicon. This means that there is a possibility that there is influence from one language to the other in terms of conceptual representation.
Differences between first language speakers and second language speakers have been found in word association research in that they make different associations to certain words, which are assumed to be caused by differences in first language associations (Verspoor, 2008). However, when asked to give the first word or words that come to mind, as is done in word association research, people have time to make conscious decisions about what they associate with a particular word. A way to test the subconscious activation of these associations is to use an experiment in which there is so little time to react that there is hardly any time for conscious processing of information. For that reason, the current study will use a priming experiment in which word associations from native speakers and word associations from language learners function as primes.
In an earlier study with a similar priming experiment (Lowie, Verspoor, & Seton, 2008; 2010), one group of students of English was tested together with a group of native speakers of English. There was a significant interaction for learners versus native speakers in highly frequent words, which showed that the Dutch students were faster on the Dutch condition, and the natives faster on the English condition. What is interesting about the current and the previous study is that the outcomes may be able to tell us something about the degree to which Dutch students of English actually become native-like speakers of English. After studying the English language at a Dutch university for four or more years, students have acquired a high command of the language and gained much knowledge about its structures and sounds. However, this does not necessarily mean that these students become fully native-like in all aspects of English. Do students of English for example acquire native-like conceptual
representations or is their own native language still a major influence? And more importantly, is it possible to detect a change in conceptual representation, moving from a Dutch representation to a more native-like representation as their proficiency also grows to be more native-like? To answer this question, different cohorts of learners of English were divided in proficiency groups and asked to perform a lexical decision task with the L1-like and L2-like2 word associations as primes to the targets.
Native speakers were used as control participants.
In the following sections I will first give an overview of the literature on conceptual representation and how this fits onto the findings of experimental research. Hereafter, I will lay out my own experiment. After the results section, I will try to place this experiment in the context of the theories and studies that were mentioned before.
2 L1 stands for first language and L2 for second language.
2 .2 .2 .
2 . T H E O R T H E O R T H E O R T H E O R E T I C A L B A C K G R O U N DE T I C A L B A C K G R O U N DE T I C A L B A C K G R O U N DE T I C A L B A C K G R O U N D
For at least tens of thousands of years people have used language to communicate, using arbitrary words to form arbitrary grammars in order to bring their meaning across. The arbitrariness arises from the fact that every lexical item could just as well be another string of words to convey the same thing. Words or lexical items are only arbitrary labels for what they represent conceptually. One could therefore come up with new words and start using them. The crucial problem is that once one person starts using new, arbitrarily formed words, communication can only take place if another person actually knows what these words mean. In order for this to happen, there needs to be some commonly shared conceptual knowledge. In other words, the difference between the words of a language and any other arbitrary words that are made up is that words of a language bear meaning in the form of conceptual representations, which is in some way shared between speakers. The starting point of the present study is the question of how such concepts and the idea of conceptual representation can be defined, and moreover, how they are similar and different between speakers.
2.1 2.1 2.1
2.1 What are concepts?What are concepts?What are concepts?What are concepts?
According to Langacker (2002), conceptualization can be anything: “a perceptual experience, a concept, a conceptual complex, an elaborate knowledge system, etc.” (p.
3). If this conceptual information can be anything, then any kind of information connected to a lexical item can be seen as constituting the conceptual representation of that item. However, as Francis (2005) argues:
[If] a researcher is interested in studying those concepts that also happen to be semantic representations of words, there is no obvious way to separate them. On the other hand, if a researcher is interested in studying concepts that are not semantic representations of words or sentences, then, ironically, it is not clear how to study them using language stimuli. (Francis, 2005, p. 252)
The conceptual representation of a certain lexical item does not necessarily consist of only verbal representations of the word. A word can be associated with for example an image, an action, or a feeling, which cannot be put into words. Because it is difficult to
test this in a behavioural language experiment, the remainder of this paper will disregard these non-verbal concepts. If restricting the idea of conceptual representation to those concepts that can actually be ‘translated’ to language, then words or chunks of words that are associated with a certain lexical items can be a reflection of the links between the concepts of these items. In terms of verbal conceptual representation, there can be many different associations with a word, which are not necessarily semantically related and/or semantic representations of the word. For example, the association of dog with the category coordinate cat, is a semantically related word, but not a representation of the word. The collocate association of speech with freedom is neither semantically related nor a semantic representation. Beside this, associative or conceptual information also includes links to phonologically or orthographically similar words. Since these different associations are difficult to tease apart (and one may wonder whether our mind differentiates between them), the present study will regard all these associations as conceptual representations.
2.2 2.2 2.2
2.2 How are concepts formed?How are concepts formed?How are concepts formed?How are concepts formed?
After defining the idea of a concept and conceptual representation, the question arises how these concepts are formed. Since concepts determine how people view the world, this question has been at the core of many ancient philosophical debates. The debate goes back to as far as Plato and Aristotle (4th Century BC) who were each taking a different perspective on where knowledge and concepts come from. Plato was an objectivist, or what we would now call a nativist, in claiming that all knowledge and representations exist, and that people only have to access what was there all the time.
Aristotle, on the other hand, was an empiricist who claimed that this knowledge could only arise through experience, since people are born as blank slates.
The present paper will regard concepts in the empiricist tradition, as being dynamic in nature, and susceptible to changes, which has also been suggested by Langacker (1987; 1998; 2000; 2002) and Lakoff (1987). Lakoff and Langacker argue for an experientialist view, claiming that conceptual structure “arises from, and is tied to, our preconceptual bodily experiences” (Lakoff, 1987, p. 267). This also applies to abstract concepts in that they have a “systematic relationship to directly meaningful structures”
(p. 268). Lakoff coined the term idealized cognitive models (ICMs) as what Verspoor (2008) has summarised as:
the dynamic model of a concept (depending on the context in which the concept is used) that includes a great deal of declarative, experiential, and image-schematic knowledge but may abstract away from everyday oddities and exceptions. We subconsciously apply such models in our everyday lives to make sense of the world around us. (p. 265)
According to this view, a concept can be defined as something that is constantly and dynamically influenced by the things we experience. Every time a certain lexical item is used, its context will have an influence on how it is perceived. Moreover, how it has been perceived in the past, might influence how it is perceived in the present, as past encounters may be activated along with the lexical item. The strength of activation depends on frequency of input and usage. In using language, the strongest activations will be activated first, and therefore come up before other activations.
In order to grasp this conceptual processing, it is necessary to understand how this is linked to what happens in the brain. Moreover, since the idea of a concept is by itself rather abstract, Langacker (1987) mentions:
Though it is customary – and I think innocuous – to use nominal expressions to designate mental phenomena (e.g. mind, concept, perception, etc.), such terms must always be understood as convenient reifications. Mind is the same as mental processing; what I call a thought is the occurrence of a complex neurological, ultimately electrochemical event; and to say that I have formed a concept is merely to note that a particular pattern of neurological activity has become established, so that functionally equivalent events can be evoked and repeated with relative ease. (Langacker, 1987, p. 100)
Thus, the formation of a concept can be explained in terms of neural processes or cognitive events (Langacker, 1987); every time a neuron is fired, this leaves a neurochemical trace, which will make it easier for neurons to travel that approximate same path again. If there is no recurrent event, the trace will fade. However, if there is recurrence, it will be facilitated by the trace. The more and more neurons travel down a particular path, the more automatized this process becomes, until it evolves into a cognitive routine. Langacker (2000; 2002) calls this process of automatization entrenchment. Conceptual representation can then be a self-organising system as it is learning and automatizing from experience. The activation of a lexical item causes conceptual representation to be activated, and thus other lexical items will be activated at the same time. Depending on the depth of the entrenchment (or the strength of activation), a lexical item can be linked to certain conventionalized concepts that make up a particular ICM. The conceptual system can be regarded as a hierarchical
representation that “takes for its starting point an integrated conception of arbitrary complexity and possibly encyclopaedic scope” (Langacker, 2002, p. 3). For that reason, this view is also known as the encyclopaedic approach (e.g. Langacker, 1998).
2.3 2.3 2.3
2.3 HoHoHoHow do concepts differ?w do concepts differ?w do concepts differ?w do concepts differ?
If conceptual representations are influenced by experience, then it means that they can differ between people. Moreover, they will differ more between people of different cultures and different languages. Sharifian (2008) mentions these differences and labels them cultural concepts. While every individual goes through different experiences and processes in forming his or her own conceptual representations which constantly change, at the same time, members of the same cultural groups can have similar experiences and therefore similar concepts that they share and through which they partially negotiate meaning. Sharifian (2008) takes this idea and proposes a culture- dependent conceptualisation, labelling these cultural differences “heterogeneously distributed cultural cognition” (p. 6). This does not necessarily mean that different individuals belonging to the same cultural group share the exact same concepts, but that they show certain overlapping representations, which are “dynamic and ever evolving” (p. 4). Sharifian also mentions that based on these overlapping representations, people make assumptions in their interaction with other people, i.e.
the expectation that the other person knows which meaning is being conveyed.
The situation becomes more complex with the acquisition of more than one language, and since the majority of people in the world are in fact multilingual, this is an issue that concerns most people in today’s world. How does the bilingual or multilingual mind manage these differing cultural concepts, and moreover, does a multilingual person distinguish between the two at all? One could ask whether a person speaking both Dutch and English distinguishes between the concept of gravy in English and the concept of jus in Dutch or whether there is a blurred or blended concept of there being a kind of watery gravy and a kind of thick gravy. Does the prototype image of a tree become both deciduous and coniferous, do these concepts exist alongside each other as two language dependent concepts, or does the most active concept of a tree (i.e. the one most encountered in the nearest history) become the prototype for tree? This issue will be discussed in the following section.
2.4 2.4 2.4
2.4 What happens to the conceptual representation of bilinWhat happens to the conceptual representation of bilinWhat happens to the conceptual representation of bilinWhat happens to the conceptual representation of bilinguals?guals?guals? guals?
Potter, So, Von Eckardt, and Feldman (1984) proposed a distinction between so-called word association and concept mediation. According to these ideas, beginning second language learners would still reach concepts of lexical items through their first language (word association), while more proficient speakers would be able to directly access the non-linguistic concepts (concept mediation). Kroll & Stewart (1994) were able to produce this pattern in an experiment with translation and picture naming. They revised the model of Potter et al. (1984) into the Revised Hierarchical Model that can be seen in Figure 1. This model accounted for translation asymmetries (translation from L2 to L1 usually being faster than L1 to L2) that were found in translation studies (e.g. Kroll
& Stewart, 1994; Sholl, Sankaranarayanan, & Kroll, 1995).
Figure 1. Revised Hierarchical Model (RHM) of Kroll & Stewart (1994). The bold lines suggest stronger links, while the dotted lines show weak connections.
This model is a static representation of how conceptual representation is accessed. The disadvantage of such a static view is that it creates an abstract and clear picture of the mind, of the way it functions with regard to the lexicon, and of the relation between lexical items and concepts. In other words, the lexicon is equated with a static dictionary and ignores the fact that the lexicon contains information and multidimensional links that are difficult, if not impossible to grasp. Because this information is missing, the static view of the lexicon as a dictionary cannot account for the acquisition of new lexical items, for the weakening and strengthening of connections between items, and for the changes in conceptualisation due to experience. In other words, it does not give us information on the continual and dynamic process that Langacker (1987) calls cognitive routines and entrenchment.
The process of entrenchment and forming cognitive routines could be elaborated in a Dynamic Systems Theory (DST) framework. Originated in the fields of
mathematics, biology, and physics, DST can be a useful theory to account for the variance in language, as it can be claimed that the process of language acquisition is so complex that it cannot be completely predictable. In line with Chaos Theory, DST
claims that it is impossible to isolate one aspect of language because it constantly interacts with other aspects and affects and is affected by these aspects as well as affecting itself. Dynamic Systems can then be defined as “a set of variables that mutually affect each other’s changes over time” (Van Geert, 1994, p. 50). All of these variables are interconnected, and “therefore changes in one variable will have an impact on all other variables that are part of the system” (De Bot, Lowie, & Verspoor, 2007, p.
8). Furthermore, from a DST perspective, language can be seen as being led by certain constants and variables. Examples of constants would be “cognitive capacity, [and]
language aptitude” and the variables would be the resources, e.g. “language competence, self-esteem and motivation” (Herdina & Jessner, 2002, p. 88). In Dynamic Systems, so-called attractor states are the near-stable states to which the development is ‘attracted’. Development will be hovering around one attractor state for some time, until it is strong enough to be moving on to the next attractor state around which it will float again for quite some time. Because different subsystems are at play here, different attractor states to each subsystem can be active at the same time.
First language acquisition and second language acquisition can be seen as processes that interact with each other, while they also form “part[s] of an overall psycholinguistic system” (Herdina et al., 2002, p. 86). In conceptual representation other variables are for example proficiency in language (which consists of different subvariables of phonology, vocabulary, pragmatics, etc.), motivation to learn the language, amount of input available, type of input available (only textual or also image or even interaction with native speakers). All these different variables interact and attain attractor states. The conceptual representation of any lexical item in the L1 or L2 can be entrenched (attracted to a state) to different degrees due to more use in one language than the other, or due to linguistic similarities between lexical items in different languages, or due to any of the variables mentioned above.
If one had to explain multilingual conceptual representation in terms of a dynamic systems perspective, someone would have a strongly entrenched L1-like representation (an attractor state) of a concept at one time, and it may take a while before a more L2- like representation would occur (another attractor state). Sub-systems could be affected by within-culture and within-subculture events, such as the influence of the title of a movie. For example, the title ‘Snakes on a Plane’ (Coane & Balota, 2008) may
cause a frequent co-occurrence that would suddenly be making the connection between snakes and planes, which could be seen as an attractor state for those particular lexical items.
A possibility of sketching conceptual representation in a dynamic distributed network can be seen in Figure 2, taken from Lowie (1998). In this Figure, change over time is visible in the t1 and the t2 figure. In the first time slot, the L2 word last is activated first for all conceptual representations of the L1 Dutch word laatste, but as the learner becomes more proficient over time, another lexical form is acquired, which can be seen in the second time slot. In the first time slot, there is complete mapping of the L1 conceptual representations on the L2 representations, whereas in the second time slot there are some conceptual meanings of the L1 word that go with one L2 lexical item, and some meanings that go with another L2 lexical item.
Figure 2. An example of the change in bilingual conceptual representation in a dynamic distributed model. LX = Lexeme, SF = Semantic Form, CR = Conceptual Representation. (Lowie, 1998)
Van Hell and De Groot (1998) came up with another visualization of shared and differing conceptual information (Figure 3). The advantage of this picture is that it accounts for overlap and for differences between lexical items and their conceptualisations, in different languages. Furthermore, the units could change from white to black, and vice versa, over time. This is a more specific, microscopic and dynamic way of looking at lexical items and their features.
Figure 3. Distributed representations of two related abstract noncognate translation pairs (upper part), two related concrete noncognate translation pairs (middle part), and two related concrete cognate translation pairs (lower part). Units are interconnected within and between subsets. (Van Hell & de Groot, 1998, p. 206).
When combining the different aspects that have been mentioned in the previous sections, it is possible to come to a converging idea of what aspects are important for the present study. The definition of a concept in this study will include that conceptual representation is constantly changing, due to both external and internal factors.
Important external factors are language and culture, which cause overall differences between two language groups. New experiences or increased/decreased use of a lexical item in a certain context may change the strength of activations or the depth of entrenchment between different items. The items with the strongest activation at one point in time will be activated first, as they are associated with the lexical item in question. Hence, since conceptual representation of lexical items is in principle an associative connection of a lexical item to the representation, conceptual information could partly be obtained from word association studies. Several researchers have conducted word association studies in order to study conceptual representation. This kind of research offers a way to find out more about conscious processing. Word associations tell us more about how lexical items are linked to others and how strong certain activations in the mind are. Although word associations are of course a representation of one point in time and not everyone will have the same associations, they can give a general idea of cultural/language-specific concepts of two different languages. The practical outcomes of research in word associations should then lead to the picture of a concept which has been described in the previous sections. These practical outcomes will be discussed in the following paragraphs.
2.5 2.5 2.5
2.5 Word association studiesWord association studiesWord association studiesWord association studies
Word association studies have been used by several researchers in the past decades in connection to conceptual information. Early monolingual word association research (e.g. Entwistle, 1966; Ervin, 2008) has shown that adults often have paradigmatic associations like synonyms and antonyms, while young children give more syntagmatic responses as, for example, collocational associations. Another type of response is a clang association, which is a phonologically or morphologically similar word. This knowledge of first-language learners has formed a basis for research in second-language acquisition. Many studies focusing on bilingual associations (e.g.
Fitzpatrick, 2006; Sheng, McGregor, & Marian, 2006; Verhallen & Schoonen, 1993) have focused their analyses on these linguistic terms. Fitzpatrick (2006) found that native speakers make more position-based associations (collocations), whereas non- native speakers make more form-based associations (comparable to clang associations).
One of the problems with this kind of analysis with classifying items according to linguistic terms is that this classification is difficult, because associations do not always clearly belong to one class, and may therefore be ambiguous. Moreover, it only looks at form and does not take into account how the meaning of the associations differs.
Van Hell and De Groot (1998) conducted a time-pressured word association task, in which Dutch students were asked to give associations in Dutch or English to Dutch or English words. In this production experiment, participants had to say out loud as quickly as possible the first word that came to their mind when seeing the stimulus word on a screen. What Van Hell and De Groot did was compare similarities in associations within and between the two languages for concreteness, cognate status and word class (verb or noun) of the stimuli. They found that cognates yielded more similar associations than non-cognates in both L1 and L2. This also applied to concrete versus abstract words (concrete words yield more similar associations) and nouns versus verbs (nouns yield more similar associations).
Another possibility in word association research would be to qualitatively analyse the differences in these associations per target word. Qualitative research into word associations has been carried out by for example Verspoor (2008). Taking a usage-based approach, Verspoor looked at the differences in associations to the word abandon by Dutch learners of English and English native speakers and the word career by Dutch and Vietnamese learners of English and native speakers of English. She found differences in associations between native speakers of English and learners of English.
For example, the Dutch learners of English had non-native-like associations of banish
and loneliness with abandon, while they did not have the native associations of ending and of giving up.
Van den Berg (2006) conducted a similar word association study for her Master thesis with three groups of advanced L2 English speakers: native speakers of Dutch, German, and French. Of the six stimuli words (house, career, time, abandon, know, and drive) she investigated, she categorized the answers, as for career: career is a type of work, career is a kind of journey, a career provides advantages, etc. She found differences between these categories for the three groups compared to the native English control group. The biggest differences were found for the abstract words, mainly the verbs. The experimental groups showed more overlap with the native Dutch, German, and French control groups (who received word association tests in Dutch, German and French respectively). For example, for the word career, Dutch and German L2 speakers of English gave Dutch- and English-like associations that had to do with how a career can provide advantages, as opposed to native speakers of English who mainly responded with words referring to the fact that “career is type of work”
(Van den Berg, 2005, p. 44). For the verb to know many participants from all three experimental groups had associations with “you can do something to know something” while this was hardly present in the native speakers of English, who had more associations with knowing things or people (Van den Berg, 2005, p. 49).
Although the groups were rather small (12 to 25 respondents in each experimental and control group, as compared to approximately 100 students in the native English control group), the study does detect a trend in conceptual transfer from the L1.
What can be concluded from these word association studies is that L1 associations play a role in L2 concepts, even when the lexical items in both languages are different from one another. However, associating words with a given lexical item is a conscious process. A way to gain information about the (multilingual) mental lexicon in a more subconscious manner is by semantic or associative priming.
2.6 2.6 2.6
2.6 Semantic or associative primingSemantic or associative primingSemantic or associative primingSemantic or associative priming and other evidence and other evidence and other evidence and other evidence
A semantic or associative priming study is a reaction time experiment that often consists of a lexical decision task with words and pseudo words as targets (meaning that the task is to decide whether the target is a word or not). These targets are preceded by words that act as a prime: if they are related to the target word, they may facilitate or inhibit a response to this target. These priming experiments try to trigger the brain
subconsciously in order to see whether certain target words are already activated by their primes and whether they are reacted to faster than target words that are not primed.
Several monolingual priming studies in the literature have pointed out differences between semantic and associative priming. Some people have argued that associative priming only taps into word form, because associations would be connected to co- occurring speech or text. Semantic priming would then be more conceptually representative, because it would tap into word meaning (Perea & Rosa, 2002).
However, studies that have tried to compare pure semantically related prime-target pairs to both associatively and semantically related pairs did not find many differences between the two (e.g. Perea et al., 2002). Furthermore, semantic distance or relatedness between different words can be measured using for example the Latent Semantic Analysis (LSA) web facility3 of Landauer and Dumais (1997), a type of analysis that generates vectors on the basis of co-occurring words in various texts. Hutchison, Balota, Cortese and Watson (2008) found that the LSA measures did not succeed in predicting priming of individual items. This suggests that there is more to priming than just semantic relatedness. The outcomes of Hutchison et al. showed that associative strength does predict priming at the item level. Word associations therefore seem to be a good representation of what connections are made in the mind.
A disadvantage of using this priming method is that there are many factors that should be controlled for in designing a study, and there are many factors that may influence the results. Hutchison et al. (2008) show that the characteristics of both the prime and the target play a role in the reaction times of a priming study. Characteristics can for example be length and frequency of both prime and target. Longer words and low frequent words yield longer response times, but also show greater priming effects (since reaction times are slower, there is more time for the prime to influence the response). The stimulus onset asynchrony also plays a role. Among other studies, Hutchison et al. (2008) found more priming in longer SOA (1250 ms) than short SOA
(250 ms). Bueno and Frenck-Mestre (2008) also reported more priming effects for longer SOAs for associative-semantic prime-target pairs as opposed to shorter SOAs and as opposed to semantic-feature prime-target pairs. On the other hand, Versace and Nevers (Versace & Nevers, 2003) found that short prime presentation (around 50 ms) would prevent any possible influence of frequency or repetition effects. Furthermore, the longer time available to process the prime, the more conscious the link between the
3 This website can be found at http://lsa.colorado.edu.
prime and target would become. In this same light, Hutchison et al. (2008) also show that forward associative strength leads to priming effects at short SOAs but backward associative strength is only a good predictor for long SOAs. This may be important for the current study.
Coane and Balota (2008) conducted a monolingual priming study with new association pairs from pop culture or politics, supported by the amount of hits of occurrences of these pairs on Google. Not only did they form these direct new association pairs, but they also mediated episodic association pairs (as mediated through the new direct associations). An example from this study would be the target word plane, with the ordinary semantic association airport, the new association snakes (from a film called ‘Snakes on a Plane’), and the mediated association cobra. In this study, even the mediated priming pairs (cobra-plane) were significantly faster than unrelated pairs. This result is in favour of the view that concepts are highly susceptible to changes in surroundings and input.
Some studies have attempted to investigate how conceptual representation of different languages can be reflected in priming experiments. For example, Dong, Gui, and MacWhinney (2005) were interested in finding out whether bilinguals store the meanings of words separately or together and how lexical processing in the L1 and the
L2 differs from each other. They tested this with third-year Chinese students of English in a lexical decision priming task with four different language conditions: Chinese- Chinese (CC), English-Chinese (EC), Chinese-English (CE) and English-English (EE).
The students received primes and targets in eight different conditions in one of the four language conditions. These primes and targets seem to have been chosen by looking at the words and coming up with expectations of differences. The experiment yielded significant results between the EC and CC condition on the one hand and CE and EE condition on the other hand (so between the Chinese targets and the English targets), but there was neither a difference between CC and EC, nor between CE and EE. It appeared that prime language did not influence the reaction times in this study.
The interpretation that the authors give of their results is that there is a distributed representation of the conceptual system, which is asymmetrical, because the reaction times to English targets were slower than the reaction times to Chinese targets.
However, it could be argued that for an asymmetrical distribution, prime language should have influenced the results as well. Furthermore, although they present their view on conceptual representation as being dynamic, their proposed model resembles a static idea of separate lexicons with one-to-one mappings to concepts. Moreover, as
these prime-target relations were not directly taken from an association task result, the question remains whether these relations were truly representative of Chinese and English associations. It would have been interesting to find out the associative strengths of their prime-target pairs.
Kotz and Elston-Güttler (2004) conducted an associative priming study using Event-Related-Potentials (ERP) to measure brain activity. They tested two groups of German participants, differing in L2 English proficiency. In their analysis, they made a distinction between associative priming (word-word) and categorical priming (word- concept). Although they did not clearly show what they meant by this distinction, they did find a significant priming effect in the reaction times for the associative primes, but not for the categorical primes in the high proficiency group, and no priming in the low proficiency group, except for a tendency towards significance in the associative condition. They reported on an earlier study they did with native speakers and claim these effects to have been similar to the high proficient group, in that they only found priming for associative primes. In language research that uses ERP, a more negative peak in the EEG wave is often found for a semantically deviating experimental item as opposed to a control item. This negativity occurs around 400 ms after the item and is usually referred to as an N400 effect. In their earlier study, Kotz and Elston- Güttler found N400 effects in the ERP data for both the associative primes and the categorical primes, whereas here they only find an N400 effect for the associative primes for both high and low proficiency groups and a small near-significant N400 effect for the high proficiency group on the categorical primes. Even though the methodology of this study is not completely clear, because they do not clearly define how they differentiate between associative priming and categorical priming, the outcomes of this experiment suggest that there are differences in conceptual representation between high and low proficiency L2 learners and that there are also differences between L2 learners and native speakers.
Lowie et al. (2008; 2010) conducted an associative priming study with a lexical decision task, which focused on conceptual transfer from L1 Dutch participants into L2 English. Using word associations of native speakers of English from a word association database, and word associations of Dutch students of English in English and Dutch students of Dutch in Dutch, prime-target pairs in English were established, which were either associates common in English or associate pairs common in Dutch (but translated - or transferred - into Dutch). A significant two-way interaction of group by prime type was found for the highly frequent target words: Dutch students of English
were faster on the Dutch-based prime-target pairs, whereas native speakers of English were faster on the English-based pairs. The outcomes of this study also suggested subconscious influence of conceptual differences between L1 and L2.
Apart from the influence of language on conceptual representation, there is in fact recent evidence that language also influences colour perception (Athanasopoulous, 2009; Thierry, Athanasopoulous, Wigget, Dering, & Kuipers, 2009; Winawer et al., 2007). These studies have shown that the difference between a light shade of blue and a dark shade of blue is perceived differently by native speakers of Greek or Russian (who have two words for these different shades of blue). This evidence seems to be solid proof for the influence of language on perception and would fit well in the theoretical explanation of how different encounters with language shape our representation of the world.
2.72.7
2.72.7 SummarySummarySummarySummary and present study and present study and present study and present study
Most of the studies that have been mentioned suggest that the conceptual representation of lexical items is dynamic and susceptible to change. Representation differs per individual, but shows overlap between people within different cultural and/or language groups (Sharifian, 2008). Moreover, word associations have shown to be qualitatively different between L1 and L2 speakers of a language (Verspoor, 2008), which predicts some cultural differences in conceptual information. This would mean that the conceptual representation of L2 learners will be influenced by concepts from their L1. If these differences in word associations are subsequently used for a reaction time experiment that relies on subconscious information processing of participants, different stages of the language learning process can be compared to find out whether there is an increase in L2 lexical concepts and whether there are any remaining links to
L1 concepts present in these L2 words. Reaction time studies using associative priming have been trying to combine word associations with the more subconscious processing of priming experiments. This has been done with either just L2-based primes (e.g. Kotz
& Elston-Güttler, 2004) with two groups of different proficiencies, or with both L1- and L2-based primes, with one proficiency group (Lowie et al., 2008; 2010). The purpose of the present study is to combine the aspects of using both L1- and L2- based primes with the testing of different proficiency groups. The study by Lowie et al.
functioned as a pilot study for the current study.
The present study will make use of a lexical decision task in a priming experiment, investigating the differences between different cohorts of L2 advanced learners of English and also comparing them to native speakers of English. Since the experiment that will be used has been adapted from Lowie et al. (2008; 2010), the results of the present study could be similar, i.e. that the cohorts will show a slight increase in recognizing conceptual relations that are native-like, which will yield faster reaction times to these specific primes when they are more advanced. However, the Stimulus Onset Asynchrony (SOA) that was used by Lowie et al. was 1250 ms. This means that the time between the presentation of the prime and the presentation of the target was more than a second. Since 1250 ms might be a process that taps into conscious processing too much, the current study uses shorter SOAs. However, this could lead to no results for backward associative strength. Since the primes were the answers to the targets in the word association tests, and they are preceding the target in the priming experiment, the items that do not have forward associative strength might not show a priming effect, as in Hutchison et al. (2008).
These issues will come back in the discussion of the results. In the next chapter, the method of the experiment will be discussed more extensively.
