Tip-of-the-tongue (TOT) states are word retrieval failures characterized by a feeling of imminent recall. TOT states can be studied as they naturally occur in everyday life using diary studies (e.g., Burke, MacKay, Worthley & Wade, Reference Burke, MacKay, Worthley and Wade1991; Gollan, Montoya & Bonanni, Reference Gollan, Montoya and Bonanni2005) or in a laboratory setting, where they are evoked by asking participants to name pictures or definitions of low-frequency words (e.g., Brown & McNeill, Reference Brown and McNeill1966). The tendency to experience TOT states seems to be sensitive to individual differences, as demonstrated by the findings of more frequent TOT states in individuals with developmental language impairment (Faust, Dimitrovsky & Davidi, Reference Faust, Dimitrovsky and Davidi1997), reading-disabled individuals (Faust, Dimitrovsky & Schacht, Reference Faust, Dimitrovsky and Schacht2003; Faust & Sharfstein-Friedman, Reference Faust and Sharfstein-Friedman2003), older adults (Burke et al., Reference Burke, MacKay, Worthley and Wade1991), bilinguals (Gollan & Brown, Reference Gollan and Brown2006; Gollan et al., Reference Gollan, Montoya and Bonanni2005), and low-proficiency second language (L2) learners (Borodkin & Faust, Reference Borodkin, Faust and Faust2012, Reference Borodkin and Faust2013, in press). The current research explored cross-linguistic transfer in the frequency of TOT occurrence. In particular, we examined whether individual differences in the tendency to experience TOT states in native language (L1) naming is related to the corresponding tendency in L2 naming, as well as to L2 connected speech production.
Cross-linguistic transfer has been described as the transfer of knowledge and skills from L1 to L2 (Cummins, Reference Cummins1981), which occurs because both languages are interdependent and rely on a common central processing system (Cummins, Reference Cummins and Bialystok1991). This effect has been observed in bilingual literacy-related skills. For instance, L1 phonological awareness was demonstrated to predict L2 phonological awareness, L2 word decoding (e.g., Durgunoglu, Nagy & Hancin-Bhatt, Reference Durgunoglu, Nagy and Hancin-Bhatt1993; Swanson, Rosston, Gerber & Solari, Reference Swanson, Rosston, Gerber and Solari2008), and L2 reading comprehension (Lindsey, Manis & Bailey, Reference Lindsey, Manis and Bailey2003).
The findings supporting the existence of cross-linguistic transfer in oral language skills (spoken and auditory) are far less conclusive and show large variability and inconsistency. For instance, L2 listening comprehension is predictable from L1 listening comprehension, but this correlation varies from r = –.04 in some samples (Proctor, August, Carlo & Snow, Reference Proctor, August, Carlo and Snow2006) to r = .67 in others (Abu-Rabia & Siegel, Reference Abu-Rabia and Siegel2002). Similar variability was observed in L1/L2 correlations when oral language skills were assessed in a narrative storytelling task, using measures such as syntactic complexity (Marchman, Martínez-Sussmann & Dale, Reference Marchman, Martínez-Sussmann and Dale2004; Pearson, Reference Pearson, Oller and Eilers2002) and lexical diversity (Kohnert, Kan & Conboy, Reference Kohnert, Kan and Conboy2010; Marchman et al., Reference Marchman, Martínez-Sussmann and Dale2004).
A recent meta-analysis of cross-linguistic transfer (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011) that included 47 studies yielded moderate-to-high meta-correlations for L1 and L2 literacy-related skills (ranging from r = .44 to r = .60). In contrast, the L1/L2 meta-correlation in oral language skills was low (r = .16). This variance in correlation magnitude has been attributed to differences in the complexity of abilities measured in each domain (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011; Proctor et al., Reference Proctor, August, Carlo and Snow2006). Within the literacy domain, L1 predictor variables are often lower-level aspects of linguistic competence (e.g., L1 phonological awareness and L1 word decoding) that involve the learning of a limited number of sounds and letters, for which a general procedure can be applied in a similar fashion for both languages, thus facilitating cross-linguistic transfer. On the other hand, L1 predictor variables, routinely examined in the oral language domain (e.g., listening comprehension), are multiply determined and thus less likely to be subjected to such common processes, which makes cross-linguistic transfer harder to detect.
According to the complexity hypothesis (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011; Proctor et al., Reference Proctor, August, Carlo and Snow2006), the lack of strong evidence supporting the existence of cross-linguistic transfer in oral language skills may be attributed to methodological issues rather than to the lack of an actual interdependence between L1 and L2 in this domain. In other words, if skills under investigation are selected with caution, it might be easier to detect cross-linguistic transfer. Elucidating the role of L1 resources available for L2 learners in L2 production may advance our understanding of the individual differences in L2 oral language skills and promote the development of teaching programs that would make better use of these resources in order to improve L2 oral communication skills.
In the current research, we tested the complexity hypothesis by examining the presence of cross-linguistic transfer in a lower-level oral language skill, namely, the ability to retrieve phonological word forms while naming, as reflected in the tendency to experience TOT states. The TOT phenomenon has been used in lexical access models as evidence that words in the mental lexicon are represented at several distinct linguistic levels, such as conceptual-semantic, syntactic, and phonological (Dell, Reference Dell1986; Levelt, Roelofs & Meyer, Reference Levelt, Roelofs and Meyer1999). The independence between these levels is apparent in TOT states, which are characterized by the speaker knowing the meaning of the intended word (i.e., it is a kind of vegetable, it is red, and it is round, for the target word tomato) and often its syntactic features (e.g., grammatical gender and form class), but being unable to retrieve its sounds (or some of them). Within lexical access models, TOT states are viewed as a success in retrieval of semantic specifications of the word, but a failure to retrieve its phonological information. This retrieval failure might arise when the phonological word forms receive insufficient activation (Burke et al., Reference Burke, MacKay, Worthley and Wade1991) or when they are underspecified, fuzzy, or imprecise (Elbro, Reference Elbro1996; Faust et al., Reference Faust, Dimitrovsky and Schacht2003).
Whether the locus of TOT states lies within the access to the phonological representations of words or their storage, it is clear that the ability to retrieve phonological word forms is a lower-level oral language skill, as it involves a basic component of speech production system (i.e., single word retrieval) and a specific level of linguistic processing (i.e., phonological). It is certainly not as complex as the skills previously measured in the oral language domain, such as listening comprehension, which involves numerous levels of linguistic processing and several components of the speech perception system, including the perceptual analysis of stress and rhythm patterns, tone patterns and intonational contours, vocabulary knowledge, and the analysis of the syntactic constituents, such as the subject, verb, and object of a sentence (J. C. Richards, Reference Richards1983). Since the retrieval of phonological word forms while engaged in L1 naming resembles more the predictor variables assessed in the literacy domain than the oral language domain in its relative lack of complexity, we expected that this ability would be correlated across languages.
The complexity hypothesis (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011; Proctor et al., Reference Proctor, August, Carlo and Snow2006) suggests that cross-linguistic transfer can be evident not only in the prediction of lower-level L2 skills, but also in more complex and multifaceted L2 skills, as long as the L1 predictor variables are sufficiently basic. This is supported by findings in literacy domain showing that L2 reading comprehension is related to L1 phonological awareness and L1 word decoding (Lindsey et al., Reference Lindsey, Manis and Bailey2003; Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011). Given this evidence, we expected that the tendency to experience TOT states in L1 naming would be associated not only with the corresponding tendency in L2 naming, but also with the more complex and multiply determined aspects of L2 oral language production, such as production of connected speech. In the current research, L2 connected speech production was evaluated in a storytelling task, using three measures: grammatical accuracy, which is the ability to produce an error-free linguistic output (Lennon, Reference Lennon1990); syntactic complexity, which refers to the range of forms in language production and the level of sophistication in these forms (Ortega, Reference Ortega2003); and lexical diversity, which reflects the range of vocabulary displayed in the output (Durán, Malvern, Richards & Chipere, Reference Durán, Malvern, Richards and Chipere2004).
The suggested relationship between TOT states and continuous language production is consistent with the predictions of sentence production models (Bock & Levelt, Reference Bock, Levelt and Gernsbacher1994; Dell, Reference Dell1986). These models agree that sentence construction, like single word retrieval, occurs at several hierarchically ordered levels of representation, starting from conceptual-semantic, syntactic, morphological, and phonological encoding, and ending with motor encoding. Although they differ in the degree of interactivity assumed among the processing levels, both serial (Bock & Levelt, Reference Bock, Levelt and Gernsbacher1994) and interactive (Dell, Reference Dell1986) models predict that phonological encoding might affect higher level encoding. Such effects are attributed to a self-monitoring mechanism within the serial models and to bottom–up activation within the interactive models. There is some experimental evidence, though surprisingly scarce, supporting these predictions (Bock, Reference Bock1987; Jaeger, Furth & Hilliard, Reference Jaeger, Furth and Hilliard2012; Lee & Gibbons, Reference Lee and Gibbons2007). In terms of sentence production models, TOT states, which interfere with phonological encoding of the planned items, may negatively affect the production of continuous language, as measured by grammatical accuracy, lexical diversity, and syntactic complexity, if the latter measures are regarded as indices of higher level encoding (i.e., morphological, syntactic, and conceptual-semantic).
Sentence production models (Bock & Levelt, Reference Bock, Levelt and Gernsbacher1994; Dell, Reference Dell1986) predict within language negative correlations between the tendency to experience TOT states and connected speech production. Accordingly, we expected that greater tendency for L2 TOT states would be related to reduced grammatical accuracy, lexical diversity and syntactic complexity in L2 narratives. More importantly to this study, which focused on cross-linguistic transfer in the oral language domain, we also expected to find evidence of correlations across languages. In line with the complexity hypothesis (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011; Proctor et al., Reference Proctor, August, Carlo and Snow2006), a positive L1/L2 correlation in the tendency to experience TOT states and negative correlations between L1 TOT states and measures of L2 continuous language production were predicted.
Methods
Participants
The sample included 50 (34 females) undergraduate students at Bar-Ilan University, aged 22.63 years on average (SD = 2.24), who were recruited through advertisements around campus. Students enrolled in an introductory psychology course received academic credit for participation. Others were paid $40 to $80, depending on the time they spent in the lab. Participants were native Hebrew speakers, who were not familiar with any language other than Hebrew and English. They have learned English as L2 in a formal school setting for about 11 years, starting at age 7, and none had an immersion experience of more than 3 months in an English-speaking country.
Materials and procedures
Participants were administered the tasks described here as part of a larger battery (Borodkin & Faust, Reference Borodkin and Faust2013, in press), to complete which they spent between 3.5 and 7.5 hours in the lab. Participants were tested in two or three separate sessions, with each session lasting up to 3 hours. A trained experimenter administered the tasks described here individually, except for Raven's Standard Progressive Matrices (Raven, Raven & Court, Reference Raven, Raven and Court1993; see below), which was given either individually or in pairs. The order of tasks administration was randomized.
Tip-of-the-tongue experiments
The participants completed a TOT naming experiment in each language. The target words used in the L1 and L2 experiments were different. The experimental stimuli were black-and-white photographs of objects and living things, which were collected through Internet search. The stimuli for each experiment were selected on the basis of the results of a pilot study that included 15 students who did not take any further part in the research. They were presented with 151 pictures to be named in Hebrew and 160 pictures to be named in English. Following the pilot study, 12 stimuli were excluded from the final list designated for the L1 experiment, because the participants were not visually familiar with the depicted object, although they knew the word (i.e., they knew what anvil means, but not what the object looks like). Eight stimuli to be named in English were omitted due to low familiarity with the target word (e.g., faucet). The final list of stimuli presented in the L1 and L2 experiments included 139 and 152 photographs, respectively.
The selection of target words was aimed at accounting for the more limited vocabulary knowledge and language use in L2 compared to L1. Consequently, the L1 list contained low-frequency words such as abacus and tripod (the full list of stimuli is available in Borodkin & Faust, Reference Borodkin and Faust2013), which might have been too difficult to name if presented in the L2 experiment and thus might have elicited too many “Don't know” responses. The L2 list, on the other hand, consisted of high-frequency words, such as knife and stairs (for a full list of stimuli, see Borodkin & Faust, in press), which might have been too easy if used in the L1 experiment and thus might have evoked too few TOT responses. Given (to the best of our knowledge) the lack of a Hebrew database that provides word frequency values directly comparable to those reported in English (e.g., Brown Corpus; Kucera & Francis, Reference Kucera and Francis1967), we confirmed the differences between L1 and L2 word lists using subjective frequency ratings. Ten participants who did not take any further part in this research were asked to rate how frequent the words in each list were in the respective language (on a scale of 1 = not frequent at all to 10 = very frequent). A t-test for independent samples yielded a significant result, t(289) = 7.14, p < .001, indicating that stimuli in L1 (M = 4.77, SD = 1.62) were less frequent than in L2 (M = 5.92, SD = 1.09). The lists also differed in word length (assessed by the number of letters), t(289) = 3.10, p = .002, such that L1 words (M = 5.14, SD = 1.46) were shorter than L2 words (M = 5.74, SD = 1.84).
The experimental procedures and the scoring criteria were the same for the L1 and L2 experiments. Both experiments were administered using the E-prime (version 1.2) software (Schneider, Eschman & Zuccolotto, Reference Schneider, Eschman and Zuccolotto2002), which was controlled by the experimenter according to participants’ responses. Pictures were presented in a random order with no time limit. Responses were collected by the program, and in some cases the records were completed manually by the experimenter. Each experiment lasted from 30 to 90 minutes.
Following the presentation of a target picture, participants could either correctly name it, provide a “Don't know” response, or report being in a TOT state, which they were told occurs when the speaker knows the word, but cannot say it at that moment. Except for pictures that triggered correct responses, all trials ended with a recognition procedure, in which participants were presented with the target word and asked if they were familiar with it. On TOT trials, participants were also asked if the target word matched the word they were searching for. An initial TOT response was scored as a positive TOT if upon the presentation of the target word participants recognized it and confirmed that it was the word they were searching for; a negative TOT was coded when the participants recognized the target word but searched for another word. When participants were unfamiliar with the target word, the initial TOT response was scored as a “Don't know” trial.
Among the various types of naming responses in a TOT experiment, positive TOT states are the most straightforward measure of the ability to retrieve phonological word forms, which was of interest here. Thus, for the purposes of the current study, only positive TOT responses were analyzed (whenever TOT states are mentioned throughout the manuscript, this type is intended). The analyses were conducted on raw and relative TOT scores, which were calculated, following Gollan and Brown (Reference Gollan and Brown2006), as the number of TOT responses divided by the sum of correct and TOT responses. Relative scores account for the number of opportunities to experience TOT states, and thus provide a way to control for differences in vocabulary knowledge (Gollan & Brown, Reference Gollan and Brown2006), which may be large in L2 learners.
Connected speech production in second language
The production of continuous language in L2 was assessed using a storytelling task, which consisted of two separate sets of sequenced pictures adapted from children books. Story 1 (Kulot, Reference Kulot2003) included 15 pictures describing a giraffe and a crocodile that fell in love and wanted to move in together. At first, they tried the crocodile's house, and when this did not work out, they moved on to the giraffe's house. In the end they built a house that was suited for them both. Story 2 (Donaldson, Reference Donaldson2001) included 14 pictures that depicted a witch who traveled riding her broom. Every now and then she had lost something, and while searching for it, she had encountered new friends, who had joined her on her journey. Towards the end of the story, the witch was attacked by a dragon and had been saved by her friends.
The participants were asked to tell each story twice, first in Hebrew and then in English. This was done to reduce inter-individual variation in the interpretation of the story plot. To ensure that the narratives were of sufficient length, participants were instructed to provide approximately three sentences for each picture. Participants required approximately 10 minutes to complete each story and were recorded while narrating in English.
The narratives were evaluated for grammatical accuracy, syntactic complexity, and lexical diversity. The scores on each of the measures were averaged across the two stories. To calculate grammatical accuracy and syntactic complexity, the data were first segmented into analysis of speech units (AS-units). An AS-unit was previously defined as “a single speaker's utterance consisting of an independent clause, or sub-clausal unit, together with any subordinate clause(s) associated with either” (Foster, Tonkyn & Wigglesworth, Reference Foster, Tonkyn and Wigglesworth2000, p. 365). The AS-unit was chosen as the basic productive unit because it reflects more reliably syntactic complexity as compared to sentence-based or clause-based units (Foster et al., Reference Foster, Tonkyn and Wigglesworth2000).
Grammatical accuracy was calculated as the ratio of error-free AS-units relative to the total number of AS-units. The equivalent measure in the written data (i.e., the mean number of error-free T-units, the latter being a minimal terminal unit) was found to be one of the most valid measures of accuracy (Wolfe-Quintero, Inagaki & Kim, Reference Wolfe-Quintero, Inagaki and Kim1998). Grammatical accuracy was scored according to the guidelines proposed by Polio (Reference Polio1997), who described six error categories. These include sentence structure errors (e.g., word order errors, such as I can't tell the all story), verb-centered errors (e.g., subject–verb agreement errors, such as They all was so happy), reference errors (e.g., quantifier–noun agreement errors, such as In this days), word-level choice (e.g., incorrect use of idiomatic phrases, as in on the second side instead of on the other hand), article errors (e.g., omission of genitives, as in So they try to live in the crocodile house), and punctuation errors. An error-free AS-unit did not contain any of the errors suggested by Polio, except for errors irrelevant in oral language production (e.g., punctuation errors).
Syntactic complexity was evaluated using the mean length of unit (MLU), calculated here as the total number of words produced during the narration divided by the total number of AS-units. The MLU is one of the most common measures employed to estimate syntactic complexity; it has adequate concurrent (Wolfe-Quintero et al., Reference Wolfe-Quintero, Inagaki and Kim1998) and predictive validity (Ortega, Reference Ortega2003).
Lexical diversity was evaluated by the D measure, which was developed by Malvern and Richards (Reference Malvern and Richards2002) to overcome problems associated with calculation of the simple type–token ratio. The type–token ratio is formed from the division of the number of types (different words) by the total number of words (tokens). Unfortunately, it varies as a function of sample length, such that longer samples tend to produce lower ratio values (B. Richards, Reference Richards1987). The D measure is based on the analysis of the probability for new words to appear in longer samples of speech or writing. This yields a mathematical model of how type–token ratio varies with token size. The D measure is obtained by comparing the mathematical model with the empirical data. Higher D values are considered to reflect higher lexical diversity. In order to obtain the D measure, the data were transferred to CHAT format and analyzed using the vocd program (MacWhinney, Reference MacWhinney2000) included in the CLAN software (MacWhinney, Reference MacWhinney1995) of the CHILDES project.
For 20% of the data, scores involving qualitative analysis (i.e., AS-units and grammatical accuracy) were coded by a second rater. The inter-rater reliability, computed using intraclass correlation coefficient, was sufficiently high for both AS-units and grammatical accuracy (.95 and .85, respectively, ps < .001).
Nonverbal intelligence
We tested the assumption that TOT responses are a specific measure of the ability to retrieve phonological word forms while naming rather than a measure of an overall intellectual ability. To this end, we evaluated the association between L1/L2 TOT states and performance on a nonverbal intelligence test, namely, Raven's Standard Progressive Matrices (Raven et al., Reference Raven, Raven and Court1993). This test involves complex visual patterns with a missing part. Below each of the patterns, six or eight segments are presented, one of which correctly completes the larger pattern. The test was administered with a 20-minute time limit. The number of correct answers was counted.
Results
Within and across language correlations
As demonstrated in Table 1, the ability to retrieve phonological word forms in naming was positively and significantly correlated across languages, in both raw and relative scores. Consistent with the cross-linguistic hypothesis, individuals with frequent TOT states in L1 naming were also more likely to experience frequent TOT states in L2 naming. Furthermore, the measures of L2 connected speech were negatively and significantly correlated within and across language with the tendency to experience TOT states, as measured in both raw and relative scores. Individuals with greater tendency for TOT states in L1/L2 naming displayed lower grammatical accuracy, syntactic complexity, and lexical diversity in L2 narratives. The linguistic meaning of these correlations is illustrated in Supplementary Materials accompanying the online version of this paper (see the journal's webpage at http://journals.cambridge.org/BIL), which provide actual narratives from two participants – one with few and the other with many L1/L2 TOT states.
Table 1. Means (M), standard deviations (SD) and intercorrelations for L1/L2 TOT states, L2 oral narrative measures, and nonverbal intelligence.
L1 = native language, L2 = second language, MLU = mean length of unit, TOT = tip-of-the-tongue
†p = .054, *p < .05, **p < .01, ***p < .001
Although the correlations between TOTs and measures of L2 connected speech production showed similar trends, regardless of TOT calculation method, they tended to be greater using relative compared to raw scores, especially for L2 TOTs (for example, the correlation with grammatical accuracy increased from r = –.39 to r = –.63 when calculated on L2 raw compared to relative TOT scores, respectively). As detailed in the Discussion below, we attribute this difference to raw TOT responses being a less reliable measure of phonological word form retrieval, particularly in L2 naming. Thus, further analyses of the data were conducted using relative scores only.
Unique contributions in prediction of connected speech production
As both across and within language correlations were significant, we also conducted multiple regression analyses predicting each of the measures of L2 continuous language production from L1 and L2 scores simultaneously. We reasoned that if both kinds of correlations mainly represent the effects of phonological encoding on higher level encoding in sentence production, than L1 and L2 TOTs will explain largely overlapping portions of variability in L2 connected speech measures. Entering L1 and L2 scores simultaneously as predictors will thus result in nonsignificant unique contributions. If, on the other hand, the correlations are affected by other factors in addition to phonological encoding (e.g., L2 proficiency might play a role in the within language correlations), than L2 and/or L1 TOT states might have significant unique contributions to variability in L2 lexical diversity, grammatical accuracy, and syntactic complexity.
For grammatical accuracy, the overall model was significant, F(2,47) = 16.17, p < .001, R2 = .41. Individuals with greater tendency for TOT states while naming in L2 had lower grammatical accuracy in L2 narratives, β = –.60, t(47) = 4.86, p < .001. The across language association between L1 TOTs and L2 grammatical accuracy was reduced and no longer statistically significant, when L2 TOT states were entered into the equation, β = –.09, t(47) = 0.72, p = .47. The multiple regression analysis on MLU scores also yielded a significant result, F(2,47) = 5.07, p = .01, R2 = .18. Individuals with greater tendency for L2 TOT states were likely to provide shorter AS-units, β = –.30, t(47) = 2.09, p = .04. After entering L2 TOT states into the equation, the across language association between L1 TOT states and MLU in L2 narratives was decreased and no longer statistically significant, β = –.20, t(47) = 1.35, p = .19. The analysis also yielded a significant result for the D measure, F(2,47) = 6.40, p = .003, R2 = .21. Individuals who experienced more TOT states in L2 naming had lower D scores in L2 narratives, β = –.39, t(47) = 2.76, p = .008. After accounting for L2 TOT states, the association between L1 TOT states and the D measure was no longer statistically significant, β = –.14, t(47) = 0.96, p = .34. Thus, regression analyses revealed that within language TOT rates contributed more to the variability in L2 connected speech than across language TOT rates.
Matching L1 and L2 lists for word frequency and length
As described above, L1 words were, on average, less frequent and shorter compared to L2 words. To evaluate the effects these differences might have had on the results described above, we re-ran the analyses after matching the lists on item difficulty. This was done by omitting the least frequent and the shortest 24 words from the L1 list and the most frequent and the longest 39 words from the L2 list. The remaining 115 L1 words and 113 L2 words did not differ significantly in frequency, t(226) = 1.89, p = .06, (M = 5.18, SD = 1.43 and M = 5.48, SD = 0.88, respectively) or length, t(226) = 1.63, p = .11 (M = 5.30, SD = 1.45 and M = 5.66, SD = 1.85, respectively). Relative TOT scores were re-calculated including only the stimuli in the matched lists and then used to re-examine the effects of cross-linguistic transfer.
The results obtained using matched and complete word lists were very similar. The tendency to experience TOT states was significantly correlated across languages, r = .44, p = .001. The across language correlations between TOT states and measures of L2 connected speech ranged from r = –.22 to r = –.33, and the corresponding within language correlations ranged from r = –.34 to r = –.63. Multiple regression analyses predicting each of the measures of L2 connected speech, based on L1 and L2 TOT scores, yielded a significant unique contribution to L2 TOT states and a non-significant contribution to L1 TOT states. These results indicate that the differences between L1 and L2 word lists in item difficulty did not affect the results observed in the analyses using complete word lists.
Tip-of-the-tongue states and nonverbal intelligence
Lastly, we calculated the correlations between L1/L2 TOT states and performance on Raven's Matrices, which, as Table 1 above demonstrates, were low and non-significant. Thus, we ruled out the possibility that TOT states reflect a general intellectual ability rather than a specific ability to retrieve phonological word forms.
Discussion
According to the complexity hypothesis (Melby-Lervåg & Lervåg, Reference Melby-Lervåg and Lervåg2011; Proctor et al., Reference Proctor, August, Carlo and Snow2006), cross-linguistic transfer in oral language skills has been difficult to identify in previous studies due to the complexity of the predictor variables applied in this domain. This hypothesis was tested in this study, which demonstrated that the ability to retrieve phonological word forms in L1 naming was related to the corresponding ability in L2 naming as well as to measures of L2 continuous language production. Individuals experiencing frequent TOT states in L1 naming were also more likely to report frequent TOT states in L2 naming and exhibited diminished grammatical accuracy, syntactic complexity, and lexical diversity while narrating in L2. Assuming that the tendency to experience TOT states is a lower-level oral language skill, which was validated by the lack of correlation between this measure and a measure of nonverbal intelligence (Raven's Matrices; Raven et al., Reference Raven, Raven and Court1993), our findings provide evidence for cross-linguistic transfer in oral language skills and support the complexity hypothesis.
The findings of this study also contribute to the under-researched topic of the phonological effects on higher level encoding in sentence production (Bock & Levelt, Reference Bock, Levelt and Gernsbacher1994; Dell, Reference Dell1986). In one of the few studies that explored such effects, inaccessibility of phonological word forms (elicited by phonological priming) was demonstrated to prompt a revision of word order in sentences (Bock, Reference Bock1987). In another study (Lee & Gibbons, Reference Lee and Gibbons2007), syntactic choice of inclusion/omission of the complementizer that in sentences like Henry knew (that) Lucy/Louise washed the dishes was found to be related to phonological preference of alternating sequence of stressed and unstressed syllables. Lastly, it was reported (Jaeger et al., Reference Jaeger, Furth and Hilliard2012) that lexical selection of a verb form (e.g., handed/passed) depended on whether the onset of the verb and the subject of the sentence phonologically overlap (e.g., Hannah handed the flask to the child vs. Hannah passed the flask to the child).
The relationship between continuous language production measures and TOT naming failures might be regarded as further evidence of phonological processing effects on higher level encoding in sentence production. Frequent TOT states, which are likely to be a manifestation of decreased availability of phonological word forms either due to access (Burke et al., Reference Burke, MacKay, Worthley and Wade1991) or storage problems (Elbro, Reference Elbro1996; Faust et al., Reference Faust, Dimitrovsky and Schacht2003), might interfere with successful completion of phonological encoding of the lexical entries chosen for production. This, in turn, might negatively affect morphological, syntactic, and conceptual-semantic encoding of a sentence. The negative correlations found between the tendency to experience TOT states in L1/L2 naming and grammatical accuracy, lexical diversity, and syntactic complexity in L2 narratives might be a reflection of such processes, if the latter measures are treated as indices of higher level encoding during sentence production. This interpretation, however, would have been more powerful, if we had been able to demonstrate L1–L1 correlations between measures of connected speech production and TOT states.
Additional factors may play a role contributing to the association between the ability to retrieve phonological word forms and connected speech production, as indicated by differences in the strength of within compared to across language correlations. L2 (relative) TOT scores seemed to be more strongly correlated with L2 grammatical accuracy, lexical diversity, and syntactic complexity compared to the corresponding L1 scores (see Table 1 above). Multiple regression analyses confirmed that TOT scores had a unique within language, but not across language, contribution to variability in the measures of L2 continuous language production. These differences remained unchanged even after partialling out L1/L2 differences in item difficulty, as was demonstrated in the analyses using stimuli lists matched on word frequency and length.
One factor that might have determined the stronger within language association was L2 proficiency. The three measures of continuous language production applied in this study, namely, grammatical accuracy, syntactic complexity, and lexical diversity, are traditionally used as measures of L2 proficiency. Furthermore, L2 proficiency has been previously observed to affect lexical access (Kroll, Michael, Tokowicz & Dufour, Reference Kroll, Michael, Tokowicz and Dufour2002). Assuming that TOT states are a measure of L2 lexical access, the stronger within than across language correlations (and the consequent greater unique contributions observed in multiple regression analyses) between TOT states and L2 continuous language production might have resulted, at least in part, from the relationship between L2 proficiency and L2 lexical access. This account, however, remains to be corroborated in future research using additional L2 proficiency measures. Although L2 grammatical accuracy, lexical diversity, and syntactic complexity have been validated against other L2 proficiency measures, such as holistic ratings and program levels (Iwashita, Brown, McNamara & O'Hagan, Reference Iwashita, Brown, McNamara and O'Hagan2008; Ortega, Reference Ortega2003), they were found to depend on factors other than L2 proficiency as well (Ortega, Reference Ortega2003; Tavakoli & Foster, Reference Tavakoli and Foster2008). A worthy alternative to these measures is the profiling procedures suggested by Processability Theory (Pienemann, Reference Pienemann1998).
In the context of measures of L2 continuous language production, the differential correlations between these measures and L1/L2 TOT states should be also noted. Specifically, L1/L2 (relative) TOT scores seemed to be correlated more strongly with grammatical accuracy than with lexical diversity or syntactic complexity (see Table 1). This difference might have resulted from the characteristics of the narrative task applied in the current study, which was based on children books. The plot in both stories was highly repetitive (e.g., in Story 2 the witch loses an item once in a while, and each time that this occurs, she lands on the ground, searches for it, and meets a new friend). Such repetitiveness might have affected measures of linguistic variability (i.e., syntactic complexity and lexical diversity) to a greater extent compared to grammatical accuracy, which might have led to less variability in these measures and consequently–to reduced correlations between these measures and TOT rates.
Our last remark concerns the effects of TOT calculation method on the strength of correlations between TOT scores and measures of L2 continuous language production. For L1 TOTs, these correlations remained largely unaltered regardless of whether calculated on raw or relative scores. For L2 TOTs, on the other hand, the correlations were higher when computed using relative compared to raw scores (see Table 1). To account for these differential effects, we considered TOT states in relation to correct responses in each language, which is illustrated using raw scores in Figure 1. As can be seen, there was a fairly consistent trend (r = –.84) in L1 naming responses, such that participants who provided many correct responses also had few TOT responses and vice versa. In other words, having many opportunities to experience TOT states often resulted in few TOT naming failures; therefore, raw and relative TOT scores provided a similar estimation of the ability to retrieve phonological word forms in L1 naming.
Figure 1. The association between TOT states and correct responses (both in raw numbers) in native language (L1; upper panel) and second language (L2; lower panel) naming.
In L2 naming, many correct responses were also associated with few TOT naming failures, but this trend was not as consistent as in L1 naming (r = –.51). As Figure 1 demonstrates, there was a group of low L2 proficiency individuals who named correctly very few L2 target words (20–60 out of 152) and reported relatively few L2 TOT states (10–20). Thus, naming in L2 – but not in L1 – in some cases elicited few TOT states, but also provided few opportunities for TOT states. In such individuals, raw TOT scores overestimated the ability to retrieve phonological word forms in L2 naming, which, as a result, led to decreased correlations between these scores and measures of L2 continuous language production. Thus, the results of this study suggest that whereas L1 naming responses are not particularly affected by the method of TOT calculation, TOT responses in L2 naming are more reliably indexed by relative than raw scores. This difference is related to greater individual variability in L2 compared to L1 vocabulary knowledge, as indirectly indicated by the number of correct responses in L1 and L2 TOT experiments.
To summarize, the present study provides evidence that the ability to retrieve phonological word forms is transferred across languages. We also demonstrated that this lower-level phonological skill can predict the more complex aspects of L2 continuous language production. These findings have a theoretical as well as practical value, as they might be helpful in developing more sophisticated L2 instruction programs, especially for individuals with difficulties in L2 language production, which aim to improve the ability to retrieve phonological word forms in L1 and L2.
