The task of reading and comprehension starts from word identification to semantic selection and integration, syntactic processing, and finally to continuous mapping of current information to the mental representation and incessant verification of the consistency across perceived messages. Many factors have a role to play on the path towards comprehension in the process, including orthographic, phonological, and morphological information, semantic richness, frequency, familiarity, plausibility, syntactic properties, and even context. The same holds true for bilingual reading.
Complex as reading already is, we can fairly claim that bilingual reading is even more complicated because two language systems may be activated at the same time, though there is only monolingual input. This non-selective view of language
activation has already been proven by numerous studies. For example, Dutch-English bilinguals were invited to participate in the experiments of Dijkstra, van Jaarsveld, and ten Brinke’s (1998). For the English lexical decision task, Dutch-English cognates, homographs, and control words that do not share such similarities were used as stimuli. Participants reacted to cognates significantly faster than to control words. In the meantime, homographs did not bring this kind of processing benefits.
This points to the facilitative effect of knowledge in a language while encountering stimuli in another language.
Dijkstra, Grainger and van Heuven (1999) conducted similar experiments to further identify how cognates, homographs, and homophones influence reading behavior. Dutch-English bilinguals were recruited, and this time the authors successfully found the impacts of each element, that is, the effects of overlap in meaning, orthography, and phonology between two languages. Significant facilitation in response time was observed for cognates and homographs, but the effects brought by homophones were the opposite, that is, participants spent significantly longer time identifying homophones than controls. According to the authors, the lack of
homograph effects in Dijkstra et al.’s (1998) paper lies in the selection of stimuli. In other words, the positive homograph effects had been confounded by the inhibitory phonological influence.
In addition, Beauvillain and Grainger (1987) also found orthographical effects by using homographs in French and English. English-French bilinguals were shown a French prime word followed by an English target word (or non-word) and had to decide whether the target was a real word. The results show that non-target language seems to have been activated initially, independent of the language context. What’s more, the frequency of each word in either language was more capable of determining whether meanings would be activated or not than the language mode per se, lending more clear support for the non-selective view of language activation.
Duyck, van Assche, Drieghe, and Hartsuiker (2007) used both a lexical decision task and a sentence-reading task to verify the effects of cognates on reading. The Dutch-English bilinguals showed a facilitation bonus on trials with cognates, in comparison with control trials, and response was significantly faster. Furthermore, even during sentence reading, cognates (both identical and non-identical) still sped up processing, shortening the response time at a statistically significance level. Though the third experiment of theirs only successfully replicated facilitation effect for identical cognates, it still attests to the idea of cross-language activation.
Also using sentence reading as the task is Schwartz and Kroll (2006).
Spanish-English bilinguals with a high- and intermediate- L2 proficiency partook in their experiments to read L2 sentences. For the highly proficient L2 group, significant cognate facilitation was observed only in low-constraint sentences, meaning there were indeed cognate effects, but the context provided by the sentence was strong enough to suppress non-selective activation (as evidenced in high-constraint
sentences). The group with less L2 proficiency generally showed similar trends.
However, one thing worthy of note is that homographs did negatively influence this second group’s performance. Homographs are identical in form but different in meaning between two languages, so this group of participants experienced
competition between meanings during the experiment, leading to higher error rates of judgment even for highly contextual sentences.
Van Hell and de Groot (2008) looked into the influence of contextual constraint of a sentence on reading as well. Cognate and concreteness effects were examined with high-constraint and low-constraint sentences. The authors invited Dutch-English bilinguals as subjects, and they were to engage in a lexical decision task or a
translation task. Participants had to first read a sentence attentively (serving as the context), and then to 1) decide whether the target word that replaced the context was a real English word or 2) translate the target word as soon as the word appeared. On the whole, cognate effects were vibrant during low-constraint sentences and largely reduced for highly contextual ones for both lexical decision and translation tasks.
Following the same line of research, Libben and Titone (2009) asked French-English bilinguals to read English sentences embedded with inter-lingual homographs and cognates in either a high- or low-semantic-constraint context.
Results again mirrored previous studies. Homographs were read significantly slower than control words under both low and high sentential constraints. Cognates, on the contrary, exhibited facilitative effects, largely shortening the reading time in both types of sentences. Specifically, this study found that non-selective access of
languages is sustained, even during early stages of reading for high-constraint sentences: The effects disappeared only in late-stage measures (i.e. 350-600 ms), while in the low-constraint context the impacts remained effective. Another interesting finding is that cognate facilitation, by and large, decreased as L2 proficiency increased, which may provide a hint towards understanding the relationship between linguistic competence and cognitive mechanisms.
In addition to orthography and meaning, phonological information in either language seems to be available for access (and even mandatory) during bilingual reading. Van Leerdam, Bosman, and de Groot (2009) asked Dutch-English bilinguals to read one English word and listen to a speech segment at a time and judge whether the two rhyme. Stimuli were all English words, with some having friendly neighbors (similar pronunciation) and enemies (different pronunciation) in English or in Dutch, some being distinct in its own way. The authors found that participants were often misled by enemy neighbors, and error rates for deciding whether the written stimulus rhymed with the speech segment rose significantly, supporting the view of
cross-language phonological activation. This mechanism seems to transcend the boundary between different alphabets because the same phenomenon has been already documented in another study, this time regarding Hebrew-English and
English-Hebrew bilinguals, by Gollan, Forster, and Frost (1997). Though asymmetric activation was observed during the experiments, the results still showed that
phonology played a part in assisting participants in reading. When reading an L2 word with an L1 prime, access to cognates (words overlapping in meaning and
phonology) was facilitated more than to non-cognates (words only overlapping in meaning), and the difference reached significance. On the other hand, the priming effects were not found with L2 primes. Interpretation and implications of this study should be treated with caution because it used loan words (cognates that are actually borrowed from English). According to the authors, this is inevitable because Hebrew
was revived as a spoken language early in the 20th century after having been used solely for religious purposes for hundreds of years. For this reason, the language lacked many lexical items for modern terminology. Although some modern words were created by adapting a Hebrew root morphologically, many modern words were simply borrowed from English (Gollan et al., 1997, p. 1123)
While loan words in Hebrew are morphologically changed, the similarity between a loan word and an English foreign word is so overwhelming that, for many people, the two words are actually identical, leading to a possible L2-L2 combination in this priming task. Luckily, a similar trend in English-Hebrew bilinguals may serve to clear this doubt.
We have seen plenty of evidence that supports the non-selective view of
language access for bilinguals. However, the aforementioned findings are confined to word-level phenomena. Does non-selectivity still hold when it comes to sentence reading and comprehension? This question is important and sound, since the topic of this current dissertation deals with sigh translation, in which sentence-level processing is as important as word recognition, if not more crucial.
Indeed, studies on bilingual sentence reading have reported findings in support of non-selectivity. What’s more, language proficiency, age of acquisition and language exposure (environment) may all have a part to play when it comes to bilingual reading (for more detail, including findings on semantic and syntactic processing in bilingual reading, see Fernández, 2002; Frenck-Mestre, 2005;
Hernández, Fernández, & Aznar-Besé, 2009).
Frenck-Mestre (1997) reported findings of several experiments, among which one instructed highly proficient French-English bilinguals to read syntactically ambiguous English sentences. The reading pattern for skilled non-native readers resembled that of native English readers. Nonetheless, the author still observed some L1 influence during the process. That is, participants hesitated when they looked at the subordinate verb in the sentences that violated lexical constraints in their native language. In another experiment, a group of beginning English-French bilinguals and a group of French native speakers both read French sentences, each with a relative clause. While the French native speakers showed preference for N1 attachment, linking the first noun phrase appearing in the sentence with the relative clause, English-French bilinguals did not have any reliable preference but a slight tendency towards N2 (the second noun phrase in the experimental sentences) attachment, which is typical in English. Though it might be possible that people tend to attach new information to the most recently processed element when faced with a not-so-familiar language, with another experiment reported in the same paper, the author was able to ascertain that the aforementioned phenomenon indeed came from L1 influence. In this
last experiment, beginning Spanish-French bilinguals were recruited to read French sentences with a relative clause. Since the Spanish participants were less skilled in French, a logical consequence following the recency attachment strategy would be that the relative clause was linked to N2 consistently. Yet, both Spanish-French bilinguals and native French participants manifested an identical inclination, which is natural in both the Spanish and French language — they all had reliable preference for N1 attachment. From Frenck-Mestre (1997), we can clearly see the impact of a bilingual’s mother tongue and how second language competence affects one’s reading pattern (see also Frenck-Mestre, 2005 for similar findings with ERPs).
Frenck-Mestre (2002) also attested to the fact that proficient second-language readers exhibit highly resembling pattern of processing with that of native readers.
According to the author, though total viewing time may be longer for L2 readers,
proficient second-language readers demonstrate an immediate sensitivity to the same factors that influence native readers’ initial progression through the sentence. Both structural ambiguity and lexical constraints influence the first pass reading times of proficient non-native readers, just as they do those of native readers (Frenck-Mestre, 2002, p. 219).
A somewhat different voice comes from Dussias (2003). Dussias invited proficient English-Spanish and Spanish-English bilinguals to fill out a questionnaire regarding choices when reading temporarily ambiguous sentences with a complex noun phrase followed by a relative clause, and to also perform a reading task to gather
real-time data. As a result, both groups preferred low attachment, similar to English monolinguals but not native Spanish speakers. The results partly contradict the notion of L1 influence. One explanation provided by the author was that the cognitive demands exerted by the stimuli might have induced the minimal effort strategies (recency attachment effect). However, an alternative and likely explanation is that language exposure plays a role, since both groups of participants live in an English environment.
The role of language exposure in reading has been further substantiated by Dussias and Sagarra (2007). Two groups of skilled Spanish-English bilinguals, living in the Spanish and the English environment respectively, and an additional group of Spanish monolinguals all read structurally ambiguous sentences as those in Dussias’s (2003) study. The results showed that monolinguals and bilinguals living in the L1 environment exhibited L1 processing preference for relative clauses; on the other hand, the group of people living in the L2 environment clearly showed L2 processing preference.
In addition to language ability and exposure, how age of acquisition affects language processing can be observed in Izura and Ellis’s (2004) translation judgment task, in trials of which participants had to decide whether the meanings of a pair of words were identical or not. Spanish-English bilinguals participated in the
experiments, and the authors found that, when translation was involved,
early-acquired L2 words were processed faster than late-acquired L2 counterparts,
while one’s native language also determined the relative speed of meaning extraction during translation.
Up to this, point, we have seen the intricate dynamic between two languages in bilingual reading. On the word level, phonological (homophones), orthographical (homographs) information and a shared origin (cognates) may affect word
identification in reading or even translation in their own way, positively or negatively.
As for factors outside the boundary of words, contextual constraints, the reader’s first language, proficiency of each language, the time when the reader learn certain words, or even the environment (language exposure in daily life) may influence the
efficiency of reading, all due to the non-selective feature of all the language subsystems in bilingual reading.