Modelling Bilingual Representation and Processing Essay

The bilingual’s ability to process several languages almost simultaneously and overcome the conflicts across languages is striking. The main topic of interest here is how a non-target language affects target word identification under various experimental circumstances. In this research, empirical investigation and computer simulation go hand in hand. To account for collected empirical data, several models of bilingual representation and processing have been developed.

This essay will examine the Bilingual Interactive Activation (BIA) model which simulates orthographic level of representation, the Distributed Feature Model (DFM) which specifies the semantics (i.e., meaning) of isolated words, and the Revised Hierarchical Model (RHM) which accounts for the interlanguage connections between lexicon and concepts/semantics as a function of L2 learners’ proficiency. The strengths and weaknesses of these three models will be evaluated on an empirical stand and the author argues that a comprehensive model (e.g., BIA+ model) is needed to simulate and account for all the above perspectives (i.e., orthographic, semantic as well as phonological representations and individual differences in terms of bilinguals’ L2 proficiency).

The BIA model (Dijkstra ; van Heuven, 1998; van Heuven, 2000) is a bilingual extension of the monolingual Interactive Activation (IA) model for visual word recognition (McClelland ; Rumelhart, 1981). In the IA model, there are three levels of nodes, with ascending complexity: (1) features of a letter such as curves, straight lines, or crossbars, (2) individual letters, and (3) words. Information at all levels can interact with each other during the word recognition process, which may flow both ‘bottom-up’ (features to letters to words) and ‘top-down’ (words to letters to features). Within levels, nodes compete for activation (thus inhibiting each other); across levels, nodes either inhibit or excite each other.

According to IA, it is these inhibitory and excitatory connections that give rise to the appropriate activation of patterns that correspond to the perception of words. As a straightforward extension of the IA model, the BIA model consists of four levels of nodes: features, letters, words, and languages. All nodes at the word level are interconnected with mutual inhibition. Feature units activate appropriate letters, and letter units activate appropriate words in the appropriate language. The model incorporates an integrated lexicon and specially adds two language nodes (one for English and one for Dutch). Each language node collects activation of all words from one lexicon and suppresses all words in the other lexicon. The relative prominence of one language or the other under particular circumstances is reflected in the activity of two language nodes.

The BIA model proposes a precise mechanism for the way in which orthographic forms are activated in two languages in bilingual visual word recognition. The model is able to simulate a series of experiments that exploit the words that share lexical features across languages. This includes cross-language neighbours that are orthographically similar in two languages but unrelated (e.g., Jared ; Kroll, 2001; van Heuven et al., 1998), interlingual homographs that share lexical form but not meaning (Dijkstra et al., 1998; Jared ; Szucs, 2002; von Studnitz ; Green, 2002) and cognates that are translation equivalents with similar spelling and sound (e.g., Dijkstra et al., 1998; van Hell ; Dijkstra, 2002). All the listed studies support the main claim of the BIA model that proficient bilinguals activate information about words in both languages in parallel. This is in line with the widely accepted account that the bilingual lexicon is integrated across languages and is accessed in a language non-selective way.

For different experiment paradigms, the model was also able to capture the effects on target recognition of a number of well-known bilingual factors. Among these are person-related factors such as language proficiency and subjective word frequency (Bijeljac-Babic et al., 1997), and contextually determined factors such as language presentation order and language of a previously presented prime (von Studnitz ; Green, 1997). However, BIA only incorporates orthographic representation and fails to fully characterize word recognition in and out of meaningful context. A complete model must specify semantics (Francis, 1999). De Groot and colleagues (de Groot, 1992, de Groot et al., 1994; van Hell ; de Groot, 1998) proposed a model of bilingual semantics, namely the Distributed Feature Model (DFM).

The model assumes that a word’s lexical category (e.g., concrete or abstract, cognates or noncognates) is determined by the degree to which semantic representations are shared (i.e., semantic overlap) across languages. More specifically, representation for concrete words and cognates are expected to be similar across languages, but not the case for abstract words and noncognates. The supportive evidence for the DFM mainly comes from translation performance as translation requires semantic processing.

It’s found by the DFM modellers that the time to recognize and produce translation equivalents is faster when the word pairs are concrete nouns and cognates because concrete words and cognates have more degrees of semantic overlap across languages than abstract words and noncognates. De Groot and her colleagues’ studies on word concreteness however, have been criticized by several researchers. They generally argue that concrete words are likely to have fewer translations across languages than abstract words. So it is not surprising that abstract words will have considerably longer translation latency due to their alternative translations (Tokowicz et al., 2002).

Neither BIA model nor DFM accounts for the developmental perspective of bilinguals in terms of their developing proficiency in their second language acquisition. To explain how the connection between words (lexical representation) and their meanings (conceptual/semantic representation) develop with increasing proficiency in the L2, Kroll & Stewart (1994) proposed the Revised Hierarchical Mode (RHM). Different from the BIA model, the RHM does not specify the precise dynamics of lexical recognition, instead, it focuses on how mappings from word to concept are developed and accessed during language processing. The model assumes independent lexical representations for words in each language but an integrated conceptual system.

There are two different pathways for the communications between the first language (L1) and second language (L2). The L1 lexicon is closely tied to an underlying conceptual memory, whereas the L2 items are mostly associated with their L1 (translation) equivalents. The conceptual memory therefore mediates the translation from L1 to L2 (forward translation), whereas direct lexical associations between L1 and L2 underlie the translation from L2 to L1 (backward translation). The bilingual memory representation described in the RHM is therefore asymmetrical. The model also assumes that as L2 proficiency increases (developmental approach), the connection between L2 words and concepts will strengthen, resulting in an increase in the strength of word-to-concept connections for the L2.

The translation asymmetry (with longer latencies in forward translation than backward translation) proposed by the RHM has been reported in several studies (e.g., Kroll & Curley, 1988; Kroll & Stewart, 1994). Kroll & Stewart (1994) reported interference effects due to the semantically categorized presentation of to-be-translated word list for forward but not backward translation. This supports the claim from RHM that backward translation does not involve conceptual processing at all. Talamas et al. (1999) tested the developmental feature of the RHM.

They examined the performance of English-Spanish bilinguals who differed in their proficiency in Spanish on translation recognition task. The critical stimuli were non-translation words that were related to the correct translation in form (similar spelling) or in meaning. Their results showed that the less-proficient bilinguals suffered more form than meaning interference, whereas the reverse was true for the more proficient bilinguals. This is in line with the developmental shift from form to meaning with increasing proficiency in L2 as illustrated by the RHM.

Among all the above 3 models (BIA, DFM, and RHM), BIA model is the only model which has been implemented computationally. A direct quantitative comparison to other models is therefore impossible. As van Heuven (2000) pointed out ‘model-to-model comparison in combination with empirical data’ will be required to fully evaluate the strengths and weaknesses of a specified model. Sunderman and Kroll (2006) however have made an admirable attempt to place the predictions of the BIA model and the RHM in the same context to investigate the lexical processing in L2.

They used the same experimental paradigm and extended design of the stimuli from Talamas et al.’s (1999) mentioned above. Their results showed that all learners experienced interference for lexical neighbours and for meaning related pairs regardless of proficiency. However, only less proficient L2 learners also suffered from form interference. The two types of lexical activation (i.e., activation of lexical form neighbours and translation equivalent) of the L1 in their study confirmed the predictions from both the BIA and RHM models.

It is now well understood that phonology of a written word plays a key role in the processes by which the word is identified (e.g., Frost, 1998; Tan ; Perfetti, 1998). The activation of a word’s phonology is presumed to be automatic and universal because all words regardless to writing systems (alphabetic, syllabic and logographic) have their constituent phonemes and pronunciation. Unfortunately, the role of phonology was neglected in bilingual visual word recognition. More recent evidence suggests that phonological codes are also active in both languages during bilingual word recogntion (Dijkstra, A. et al., 1999; Jared & Kroll, 2001; Jared & Szucs, 2002). None of the models examined above accounts for the existence of this phonological activation.

The BIA model has been extended as BIA+ model (Dijkstra, 2005; Dijsktra & van Heuven, 2002; Thomas & van Heuven, 2005) to include orthographical, as well as phonological and semantic representations in the word identification system, and a distinction is made between a word identification system and a task/decision system. The inclusion of orthographic, phonological and semantic representations implies that word recognition can be affected by spelling, sound and meaning of a word. The BIA+ model proposes that these three codes are interactive and bi-directional, therefore activation flows from orthographic codes to phonological and semantic codes (feedforward activation), and also feedback from semantic and phonological codes to orthography (feedback activation).

However, this bi-directional flow of activation among the three codes and the role of feedback activation are open for much debate even in the monolingual literature. The issue mainly concerns the inconsistencies in the mappings between phonology and orthography (Stone, Vanhoy and van Orden, 1997), and between semantics and orthography (Pecher, 2001). Van Hell (2002) pointed out that in terms of the discussion of semantic presentation in the identification system of the BIA+ model, the modellers only focused on their findings on interlingual homographs and cognates and assumed that interlingual homographs do not share meaning whereas cognates do. However, some studies suggest that cross-linguistic semantic overlap is complicated and the empirical data reported are not always consistent.

For example, van Hell and de Groot (1998) asked Dutch-English bilinguals to associate to the same stimuli both in the language of the stimuli (within-language) and in the other language (between-language). They found that the within- and between-language associations for concrete cognates were more often translations of one another than those for abstract cognates. Moreover, they reported that bilinguals found it easier to retrieve an association for concrete cognates than for abstract cognates. They therefore concluded that cognates can differ in their degree of semantic code overlap. Implementing semantic representations in the BIA+ model therefore appears to pose problems.

The BIA+ model is open for much critique in different perspectives. Green (2002) questioned the scope of the BIA+ model by concerning the relationship between the BIA+ (cognitive) model and neuroimaging data. He also suggested the BIA+ model to take into account the individual differences in bilingual visual word recognition (see also van Hell, 2002). Thomas (2002) sensed that the BIA+ model is ‘chasing the empirical data rather than predicting it’ due to the fact that the model has been adapted to account for recent empirical data.

Jacquet and French (2002) spotted the lack of learning mechanisms being the most significant problem with the BIA+ model. Li (2002) pointed out that there are so many variables that affect bilingual word recognition that one hardly could take everything into account to develop a formalized model. In response to these critiques, Dijkstra and van Heuven (2002) proposed further directions for future research. Despite the fact that the development of the BIA+ model is not yet complete, this inspiring model has certainly opened new approaches to the structure and dynamics of bilingual word recognition. We are eager to see the further development of the BIA+ model and anticipate that the model construction will take into account all the perspectives with existing models discussed in this essay.

To conclude, all the existing models (BIA, DFM and RHM) provided us with different approaches to bilingual representation and processing. However, they are open for much critique due to their inability to account for all the variables/factors that affect bilingual word recognition. The BIA+ model has made a step towards a more complete formalization. A completed computational implementation of BIA+ model will enhance our understanding of bilingual language processing.

References:

Bijeljac-Babic, R., Biardeau, A., & Grainger, J. (1997). Masked orthographic prming in bilingual word recognition. Memory & Cognition, 25, 447-457.

De Groot, A. M. B. (1992). Determinants of word translation. Journal o f Experimental Psychology: Learning, Memory and Cognition, 18, 1001-1018.

De Groot, A. M. B., Dannenburg, L., & van hell, J. G. (1994). Forward and backward word translation by bilinguals. Journal of Memory and Language, 33, 600-629.

Dijkstra, A. (Ton) (2005). Bilingual visual word recognition and lexical access. In J. F. Kroll & A. M. B. de Groot (Eds.), Handbook of Bilingualism: Psycholinguistic Approaches (pp. 179-201). Oxford: Oxford University Press.

Dijkstra, A. (Ton), Grainger, J., & van Heuven, W. J. B. (1999). Recognition of cognates and interlingual homographs: The neglected role of phonology. Journal of Memory and Language, 41, 496-518.

Dijkstra, A. (Ton), & van Heuven, W. J. B. (1998). The BIA model and bilingual word recognition. In J. Grainger & A. M. Jacobs (Eds.), Localist Connectionist Approaches to Human Cognition (pp. 189-255). Mahwah, NJ: Erlbaum.

Dijkstra, A. (Ton), & van Heuven, W. J. B. (2002). The artchitecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition, 5(3), 175-197.

Dijkstra, A. (Ton), van Heuven, W. J. B., & Grainger, J. (1998). Simulating cross-language competitor effects with the Bilingual Interactive Activation Model. Psychologica Belgica, 38, 177-196.

Francis, W. S. (1999). Cognitive integration of language and memory in bilinguals: Semantic representation. Psychological Bulletin, 125, 193-222.

Frost, R. (1998). Toward a strong phonological theory of visual word recognition: True issues and false trails. Psychological Bulletin, 123, 71-99.

Grainger, J. (1993). Visual word recognition in bilinguals. In R. Schreuder & B. Weltens (Eds.), The Bilingual Lexicon (pp. 11-26). Amsterdam/Philadelphia: John Benjamins.

Green, D. W. (2002). The bilingual as an adaptive system. Bilingualism: Language and Cognition, 5(3), 206-208.

Jacquet, M., & French, R. M. (2002). The BIA++: Extending the BIA+ to a dynamical distributed connectionist framework. Bilingualism: Language and Cognition, 5(3), 202-205.

Jared, D., & Kroll, J. F. (2001). Do bilinguals activate phonological representations in one or both of their languages when naming words? Journal of Memory and Language, 44, 2-31.

Jared, D., & Szucs, C. (2002). Phonological activation in bilinguals: Evidence from interlingual homograph naming. Bilingualism: Language and Cognition, 5, 225-239.

Kroll, J. F., & Curely, J. (1988). Lexical memory in novice bilinguals: The role of concepts in retrieving second language words. In M. Gruneberg, P. Morris & R. Sykes (Eds.), Practical Aspects of Memory (Vol. 2, pp. 389-395). London: John Wiley & Sons.

Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33, 149-174.

Li, P. (2002). Bilingualism is in dire need of formal models. Bilingualism: Language and Cognition, 5(3), 213.

McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception. Part 1: An account of basic findings. Psychological Review, 88, 375-405.

Pecher, D. (2001). Perception is a two-way junction: Feedback semantics in word recognition. Psychonomic Bulletin & Review, 8, 545-551.

Stone, G. O., Vanhoy, M., & van Orden, G. C. (1997). Perception is a two-way street: Feedforward and feedback phonology in visual word recognition. Journal of Memory and Language, 36, 337-359.

Sunderman, G., & Kroll, J. F. (2006). First language activation during second language lexical processing: An investigation of lexical form, meaning and grammatical class. Studies in Second Language Acquisition, 28, 387-422.

Talamas, A., Kroll, J. F., & Dufour, R. (1999). Form related errors in second language learning: A preliminary stage in the acquisition of L2 vocabulary. Bilingualism: Language and Cognition, 2, 45-58.

Tan, L. H., & Perfetti, C. A. (1998). Phonological codes as early sources of constraint in Chinese word identification: A review of current discoveries and theoretical accounts. Reading and Writing, 10, 165-200.

Thomas, M. S. C. (2002). Theories that develop. Bilingualism: Language and Cognition, 5(3), 216-217.

Thomas, M. S. C., & van Heuven, W. J. B. (2005). Conputational models of bilingual comprehension. IN J. F. Kroll & A. M. B. de Groot (Eds.), Handbook of Bilingualism: Psycholinguistic Approaches (pp. 202-225). Oxford: Oxford University Press.

Tokowicz, N., Kroll, J. F., de Groot, A. M. B., & van Hell, J. G. (2002). Number of translation norms for Dutch-English translation pairs: A new tool for examining language production. Behaviour Research Methods, Instruments, & Computers, 34, 435-451.

van Hell, J. G. (2002). Bilingual word recognition beyond orthography: On meaning, linguistic context and individual differences. Bilingualism: Language and Cognition, 5(3), 209-212.

Van Hell, J. G., & de Groot, A. M. B. (1998). Conceptual representation in bilingual memory: Effects of concreteness and cognate status in word association. Bilingualism: Language and Cognition, 1, 193-211.

Van Hell, J. G., & Dijkstra, A. (Ton) (2002). Foreign language knowledge can influence native language performance in exclusively native contexts. Psychonomic Bulletin & Review, 9, 780-789.

van Heuven, W. J. B., Dijkstra, A. (Ton) & Grainger, J. (1998). Orthographic neighbourhood effects in bilingual word recognition. Journal of Memory and Language, 39, 458-483.

van Heuven, W. J. B. (2000). Visual word recognition in monolingual and bilingual readers: Experiments and computational modelling. Ph.D. thesis, NICI, University of Nijmegen, Netherlands.

von Studnitz, R. E., & Green, D. W. (1997). Lexical decision and language switching. International Journal of Bilingualism, 1, 3-24.

von Studnitz, R. E., & Green, D. W. (2002). Interlingual homograph interference in German-English bilinguals: Its modulation and locus of control. Bilingualism: Language and Cognition, 51, 1-23.