This study examines the impact of the
crosslinguistic similarity of translation equivalents on
word recognition by Russian-English bilinguals,
who are fluent in languages with two different
but partially overlapping writing systems.
Current models for bilingual word recognition, like
BIA+, hold that all words that are similar to the
input letter string are activated and considered
for selection, irrespective of the language to
which they belong
Many Russian-English translation equivalents are in part composed of shared letters that can potentially activate both Russian and English word candidates. Often, these letters have ambiguous phonemic mappings across the two languages. The degree of ambiguity is high especially when shapes of block-letters and letters in italics overlap across languages. For instance, a printed Russian letter ‘и’ does not look like any letter of the English alphabet, but the shape of its handwritten equivalent ‘u’ perfectly coincides with the English hand-written grapheme. We identified 5 overlapping pairs of printed English block-letters and Russian letters in italics (g, r, m, n, u).
Our study started from the assumption that
even when a bilingual reads English words in
printed font, letter shapes also activate
handwritten Russian letters with similar shapes in a
bottom-up way. We focused on the impact of
convergence and divergence in Russian and English
script coding for cognates and non-cognates.
Cognates are translation equivalents with
significant cross-linguistic form overlap in phonology
and/or orthography (e.g., ‘marriage’ in English,
‘mariage’ in French). Cognates are generally
processed more quickly by bilinguals than
matched control words
We are making the following predictions about
English word recognition by Russian-English
bilinguals:
1. In English word processing, Russian-English
bilinguals will activate lexical candidates that are
similar to the input word in both Russian and
English (language non-selective lexical access).
2. English-Russian cognates will be recognized
more quickly than English control words, due to
co-activation and convergence
The following two predictions are more speculative and exploratory in nature. 4. Response times to cognates with transparent orthography, i.e. shared letters mapping onto the same phonemes in the two languages, will be about equal to those for cognates with mismatching orthography, because transparent orthography and shared phonology will lead to increased lexical competition, but, at the same time, the transparency will lead to increased semantic coactivation of cognates in the two languages. 5. English control words with mismatching orthography will be processed more quickly than words with ambiguous orthography, because less interference from the Russian alphabet is expected in the first case. 3
To test these hypotheses, we first constructed a large database of Russian-English cognates with three, four, five or six letters in length. To our knowledge, no such database is currently available to the community of researchers. Next, 75 English cognates were selected as test words in a lexical decision task. Orthographic coding was performed on English cognate words written in lower-case block letters in Arial font. The resulting items were allocated to three categories: 1) Cognates with Ambiguous Orthography (CAO=Minus condition), composed of letters that have different phonological mappings in English and Russian (e.g. ‘guru’ might be read as /digi/ if a Russian monolingual was asked to read this string of letters); 2) Cognates with Transparent Orthography (CTO=Positive condition), composed of letters that largely share their orthographic-phonological mappings with letters of the Russian alphabet (e.g. in ‘koala’ the only mismatch with the Russian alphabet is the grapheme ‘l’); 3) Cognates with Mismatching Orthography (CMO=Base condition), composed mostly of letters that do not exist in the Russian alphabet (e.g. ‘filter’). The cognate types were matched across conditions (CAO/CTO/CMO) in word length, frequency, and degree of cross-linguistic orthographic overlap between Russian and English alphabets. Three groups of control words were then selected that matched the cognates of each type with respect to these three dimensions. Finally, each cognate and non-cognate was matched with a pseudo-word generated with the help of the Wuggy-software (crr.ugent.be).
Next, 20 Russian-English bilinguals were asked to rate the visual similarity between the English cognates and their Russian translation equivalents. They also rated the semantic similarity of all selected item pairs. Rating results showed that bilinguals mostly considered orthographic congruence (as opposed to incongruence) between the orthography of Russian and English translation equivalents and gave higher ratings to English words that have shared orthography with the Russian alphabet. Ratings also indicated that bilinguals considered not only block-letters but also corresponding handwritten graphemes when rating the visual similarity between words.
In total, 37 Russian-English participants (10 male vs. 27 female; age: 19-60 years) took part in the study. At the moment of testing, all participants were residing in English-speaking countries: 11 participants in Bristol, UK, 21 participants in Sheffield, UK, and 5 participants in New Zealand. After the experiment, all participants rated their proficiency in English on a scale from 1 (the lowest) to 6 (the highest). Average ratings for reading, writing, speaking, and listening varied between 4.4 and 5. Except for two participants, ages of L2 acquisition (AoA) ranged between 6 and 19 years. Length of residence in an English-speaking country varied between 3 months and 21 years (mean = 33 years, SD = 11 years).
Participants performed an English lexical decision task, in which they pressed a “yes” or a “no” button depending on whether a presented word was English or not. They were asked to press a button as quickly and accurately as possible. The items were presented in a pseudorandomized order to each participant. The experiment was programmed in E-Prime. Reaction times (RTs) and accuracy of responses were measured. Only correct responses to real words were included in the analyses of reaction times. 4
First, all responses faster than 300 ms and slower than 3 s were removed from the data set, because they were not considered as valid measurements. Next, the data from 9 participants were excluded from analysis, because they had a response accuracy below 70%. We removed 5 cognates, 8 control words, and 14 non-words from the items, because these items had an accuracy below 70 % or had extremely slow responses. For the remaining 28 participants, after removing these items, cognate and control word conditions were still matched with respect to length and frequency (as shown by non-significant t-tests). None of the remaining responses were further apart than 2.5 SDs from the participant mean in each condition. The mean RT for non-words was 892 ms. Table 1 presents the mean RTs for words in each cognate and control word condition, as well as their accuracy.
repeated-measures Analysis of Variance (ANOVA), using cognate type (3, MO vs. AO vs. TO) and cognate status (2, cognate vs. control) as within-subject factors. This analysis resulted in main effects of Cognate Status (F (1, 27) = 94.11, p<.001), Item Type (F (2, 54) = 9.89, p<.001), and an interaction of Cognate Status with Item Type (F (2, 54) = 10.22, p<.001). Next, we did planned comparisons to test the Cognate Minus (CMO) and Cognate Plus (CTO) conditions against the Cognate Base (CMO) condition. Significant differences were found between the RTs between the Cognate Base condition and the Cognate Minus condition (t(27)=5.0, p<.001 two-tailed) but not between the Cognate Base and the Cognate Plus condition (t(27)=.60, p=.55). There was a significant difference between the Cognate Base condition and the Control Base condition (t(27)=-6.54, p<.001). Finally, no significant differences arose between the different control conditions (Control Base vs. Control Minus, t(27)=-.67, p=.51; Control Base vs. Control Plus t(27)=-.36, p=.72). 5
Russian-English bilinguals performed an English lexical decision task with purely English control words and English-Russian cognates 1) with mismatching orthography or 2) shared orthography with a) transparent or b) ambiguous mappings on phonemes in Russian and English.
Responses to cognates were faster than to English controls (see Table 1). This cognate facilitation effect is in line with prediction 1 that lexical candidates in both Russian and English are activated during Russian-English bilingual word recognition.
It also confirms prediction 2 that language
non-selective lexical access takes place in
Russian-English word recognition. Because the
effect is also observed in cognates with (partially)
mismatching orthography, the cognate effect
may in part be ascribed to the phonological and
semantic overlap in these cognates. Thus, the
orthographic input representation quickly leads
to an activation of sublexical and lexical
phonological representations
In line with prediction 3, the cognate facilitation effect is modulated by the degree of shared transparent overlap between Russian and English alphabets. Cognates with transparent orthography were processed faster than cognates with ambiguous grapheme to phoneme mappings. This finding can be explained by assuming that Russian words are co-activated with English words to the extent that they match the English letter input, irrespective of whether this matching is in terms of block letters or handwritten visual similarity. In other words, it is purely a bottomup (signal-driven) effect.
The finding that cognates with mismatching
orthography and shared orthography with
transparent grapheme-to-phoneme mappings are
responded to about equally fast, is in line with
prediction 4, which is based on the representation
for cognates that has been proposed by Dijkstra,
Miwa et al. (2010). As Figure 1 indicates, both
form representations of cognates are assumed to
be activated based on the input and they spread
activation to convergent semantic
representations. The co-activation of form representations
results in lexical competition and interference
Finally, in contrast to prediction 5, English
control words with mismatching orthography
were not processed more quickly than control
words with ambiguous orthography. Apparently,
mismatching orthography in general did not
result in any systematic interference on word
processing speed. Said differently, the noise
introduced by spuriously activated word candidates
from Russian with overlapping letters in the
other control conditions did not systematically affect
the lexical decision to the English target word,
although it may have affected the participants’
general decision-making strategies in the
experiment. In terms of interactive activation
models, the increase in noise could be cancelled out
by a somewhat higher reliance on semantic codes
or global lexical activation
In all, the obtained patterns of results are in
support of interactive activation models for
bilingual word recognition, such as the BIA+
model
This study confirms the presence of language non-selective lexical access in visual word recognition by different script-bilinguals, in line with, e.g., for Korean-English Kim and Davis (2003) and for Japanese-English Hoshino and Kroll (2008), Miwa et al. (2014), and Ando et al. (2015). Moreover, it bridges research on shared scripts and different scripts by considering the partially overlapping Latin and Cyrillic scripts of English and Russian. It is innovative in showing that cross-linguistic effects depend on the degree of overlap in scripts depending on the exact characteristics of the words involved.
The study also provides indirect support for
various types of models that assume
coactivation of word candidates that are
orthographically similar to the input letter string. The
set of such candidates is often referred to as the
neighbourhood
To conclude, we presented evidence in favor of language non-selective lexical access in Russian-English bilinguals, showing an EnglishRussian cognate facilitation effect, the size of which depended on whether there was overlap in orthography or not, and on whether this overlap was ambiguous or transparent relative to phonology. These effects were shown to be lexical in nature, because mismatching orthography in control target words with translations that are completely different in form did not show any evidence of differential processing.
This research was made possible with support from NetWordS, the «European Network on word structure in the languages in Europe» (research grant n° 09-RNP-089). The authors are also deeply indebted to the EPSRC's RefNet research network that enabled us to collect a large part of our data. We also wanted to thank the anonymous reviewers for their helpful comments on an earlier version of this paper.