BACKGROUND

Recent studies have provided evidence that the availability of written forms in the L2 input can aid learners’ memory for the phonological forms of newly learned words. These studies have explored cases where learners who are exposed to words’ written forms exhibit more accurate memory for the phonological forms of words than do learners who do not see written forms. For example, Escudero et al. (Reference Escudero, Hayes-Harb and Mitterer2008) found that the availability of written forms supported native Dutch speakers’ ability to distinguish the phonological forms of newly learned English (non)words in lexical memory. Native Dutch speakers have difficulty perceiving and producing the English /æ/–/ε/ contrast (e.g., “bat” vs. “bet”; see, e.g., Weber & Cutler, Reference Weber and Cutler2004). However, Escudero et al. hypothesized that when new auditory words contrasted by /æ/ and /ɛ/ were accompanied by their written forms (e.g., “tenzer” for [tɛn . . .] and “tandek” for [tæn . . .]) in the input to Dutch learners of English, this would facilitate the Dutch speakers’ memories for the words’ phonological forms. In the study, one group of participants saw spelled English nonwords and heard auditory forms, while the other group only heard the forms. Patterns of lexical confusion revealed by eye-tracking data indicated that only the participants exposed to the orthographic forms had created lexical representations that contrasted the /æ/ and /ɛ/ vowels in the first syllables of the newly learned words. Thus, orthographic information in the study facilitated participants’ ability to encode the novel contrast in their lexical representations.

There are cases where orthographic information may be neither a help nor a hindrance to learners. For example, Simon et al. (Reference Simon, Chambless and Alves2010) examined whether orthographic information would support learners’ ability to acquire a novel phonological contrast. The French vowel contrast /u/–/y/ is notoriously difficult for English speakers, who lack the high, front vowel /y/. Participants in the orthographic and auditory information group saw pictures (e.g., a boat, a banana, and glasses) and heard triplet auditory forms (e.g., [styg]–[stug]–[stig]) while seeing the corresponding orthographic representations (e.g., <stûgue>–<stougue>–<stigue>). It was found that orthographic information only minimally aided participants, with performance not significantly above the auditory only group. Thus, in some cases, orthographic information may not affect learners’ ability to remember the phonological forms of words (a conclusion that is supported by the findings of Escudero, Reference Escudero2015 [this issue]).

Finally, there is also evidence that written input can interfere with learners’ lexical–phonological development, specifically in cases where the first and the second language use the same set of graphemes but employ different grapheme–phoneme correspondence rules. Hayes-Harb et al. (Reference Hayes-Harb, Nicol and Barker2010) found that when the spelled forms of words presented during a word-learning phase did not follow English grapheme–phoneme correspondences (e.g., the auditory word [togɛg], associated with a picture of a button was written as <thogeg>), native English speakers misremembered the phonological forms of the words according to their spellings (e.g., at test, they incorrectly accepted [Ɵogɛg] as matching the picture of the button). The results of Hayes-Harb et al. suggest that learners who interpret grapheme–phoneme correspondences in novel words following first language (L1) correspondences may remember the novel phonological forms incorrectly. Similar effects of mismatches between native and L2 spelling conventions for native speakers of Italian learning English are reported by Bassetti and Atkinson (Reference Bassetti and Atkinson2015 [this issue]).

Bassetti (Reference Bassetti2006) demonstrated that a learner's L1 spelling conventions can hinder acquisition of phonological forms in the L2. She found that native English speakers learning Chinese interpreted Pinyin, a Roman alphabet representation of Mandarin, using L1 grapheme–phoneme correspondences. When asked to segment and to count diphthongs and triphthongs in Mandarin listening tasks (e.g., for <gui>–[guei] and <wei>–[uei]), native English-speaking learners of Mandarin in this study counted and segmented fewer vowels than were present (i.e., [gui]), presumably due to inferences they made about the phonological forms of the utterances based on how they were spelled in Pinyin.

The studies discussed thus far have considered cases where learners encounter L2 written forms in a familiar writing system. Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2013) investigated whether novel written information (in this case, diacritic tone marks) can similarly influence learners’ memory for the phonological forms of newly learned L2 words. One group of participants heard auditory forms and saw segment sequences in pseudo-Mandarin with diacritic marks representing the four-tone contrast in Mandarin (e.g., <fián>), while another group received orthographic information without tone diacritics (e.g., <fian>). The final test assessed whether participants had made lexical representations between the auditory forms and pictures. Participants in the tone marks condition performed more accurately, suggesting that the availability of the diacritics had facilitated their memory for the phonological forms of words. Showalter and Hayes-Harb demonstrated that learners are able to make use of novel orthographic information, in this case, diacritic information.

The studies discussed thus far have provided evidence that learners can make inferences about the phonological forms of newly learned L2 words based on information provided by written forms, and that these inferences are sometimes helpful and sometimes not. Why is orthographic input helpful in some cases and not in others? One possibility is that the auditory contrasts explored in the various studies may differ in their perceptibility to subjects. For example, the novel vowel contrasts in the Simon et al. (Reference Simon, Chambless and Alves2010) study may have been so difficult that the orthographic input was ultimately unusable by subjects. This issue will be taken up in more detail in the Discussion section.

Another possibility is that a learner's ability to benefit from the availability of L2 written input, at least when the L1 and L2 employ the same writing system, may have to do with the relationship between the relevant grapheme–phoneme correspondences in the L1 and the L2 (i.e., whether they are “congruent” or “incongruent”). Hayes-Harb et al. (Reference Hayes-Harb, Nicol and Barker2010) found that incongruency between the L1 and L2 grapheme–phoneme correspondences served to interfere with their subjects’ ability to remember the phonological forms of words; Escudero, Simon, and Mulak (Reference Escudero, Simon and Mulak2014) report a similar effect of incongruency with native Spanish learners of Dutch, and Young-Scholten and Langer (Reference Young-Scholten and Langer2015 [this issue]) provide evidence that this orthographic effect also occurs in production of the L2 (e.g., German <s>–[z] by English speakers). In contrast, Escudero et al.'s (Reference Escudero, Hayes-Harb and Mitterer2008) subjects presumably benefitted from the relative congruency between English and Dutch grapheme–phoneme correspondences (i.e., the letters “a” and “e” map to separate vowel phonemes in both languages). Rafat (Reference Rafat2015 [this issue]) reports a similar effect for native speakers of English and Spanish rhotics, where the availability of spelled forms (with rhotics represented by <r>) appears to increase the likelihood that subjects produce the corresponding segments as rhotics (as opposed to, e.g., fricatives).

To the extent that “incongruency” between grapheme–phoneme correspondences in the L1 and the L2 may interfere with L2 word-form learning, difficulty associated with novel graphemic forms in the input may be offset by the fact that there are no native grapheme–phoneme correspondences to “override” in cases where the L2 uses a different writing system. This hypothesis is supported by the finding by Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2013) that native English speakers experienced a benefit from the availability of diacritic tone marks in learning to distinguish among lexical tone minimal quadruplets. Thus, when the majority of the orthographic symbols are familiar and novel graphemic forms cue a novel phonological contrast, learners may benefit from the availability of written forms. In the present study, we ask, can learners benefit from written forms made up of entirely unfamiliar symbols? To address this question, we examined native English speakers’ acquisition of Arabic-like words differentiated by the voiceless velar–uvular stop contrast and written in the Arabic script.

EXPERIMENT 1

Experiment 1 was designed to determine whether native English speakers can benefit from the availability of written input in an entirely novel script when learning the phonological forms of new L2 words. Previous studies have primarily focused on orthographic input cases where the graphemes and, for the most part, the phonemes were familiar in some way to the learners. To investigate whether learners are able to invoke a more abstract type of knowledge, that is, the expectation that written forms will provide useful information about the phonological forms of words, we selected the Arabic alphabet, which is entirely different from the Roman alphabet. The Arabic alphabet differs from the Roman alphabet as used in English with respect to the inventory of phonemes represented by the writing system, the letters themselves, the representation of vowel phonemes (i.e., short vowels are indicated via diacritics if at all), and the directionality of the script (Arabic is written right to left). In order to maximize consistency with previous studies (e.g., Escudero et al., Reference Escudero, Hayes-Harb and Mitterer2008; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2013), we neither directed participants’ attention to the written forms nor instructed them in any way about the Arabic script.

Stimuli

All stimuli in Experiment 1 were Arabic nonwords created in a C₁VC₂V template, three of which were of the form C₁V₁C₂V₂ and three of the form C₁V₁C₂V₁ (e.g., [kita], [kini]). There were 6 minimal pairs, contrasted by the velar–uvular contrast /k/–/q/ (e.g., [kubu]–[qubu]), for a total of 12 unique words. The contrasting segments /k/ and /q/ appear in word-initial position, which was chosen for its perceptual salience. All vowels are long (i.e., /a:, u:, i:/) and thus were represented in the written forms via letters (as opposed to diacritics).

Two male talkers of Jordanian Arabic (ages 30 and 21) produced the nonwords. The talkers had been in the United States for nearly equivalent lengths of time (2 and 2.5 years). The talkers were shown the nonwords written in Arabic and were asked to produce each one three times. The second production of each nonword was extracted from the recordings for use in the present study.

Each nonword was randomly assigned an object picture (from the Bank of Standardized Stimuli; Brodeur, Dionne-Dostie, Montreuil, & Lepage, Reference Brodeur, Dionne-Dostie, Montreuil and Lepage2010; Creative Commons, n.d.); this assignment was identical for all subjects. Table 1 provides the orthographic representations, auditory forms, and the pictured meaning for each nonword.

Table 1.

Experiment 1 auditory and visual stimuli

Procedure

Experiment 1 consisted of a word-learning phase, a criterion test phase, and a final test phase. All phases of the experiment were conducted within an hour-long session. The word-learning phase for the Arabic script condition participants involved simultaneously hearing the auditory form of each word, seeing the picture, and seeing the written form (e.g., hear [kubu], see a picture of glasses, and see the orthographic representation). The Arabic script condition participants saw the words written in the Arabic script (e.g., ), and the control condition saw the meaningless Arabic letter sequence <>, roughly analogous to <XXXX>, used as a control form in studies of Roman orthography. Use of this control written form ensured that participants in both groups experienced similar amounts of visual information.

In both word-learning conditions, each of the auditory words, along with its corresponding picture and written form, was presented twice per block, totaling 24 items per block. The block was presented four times in a different, random order each time and for each participant, for a total of 96 exposures to the new words (12 words × 2 presentations/block × 4 blocks). Participants were not required to register any responses during the word-learning phase; they were simply instructed to learn the words and their meanings as well as possible.

The purpose of the criterion test was to assess whether the participants had generally learned the word meanings. Each trial of the criterion test involved the presentation of an auditory word and a picture without a written form (see Figure 1 for examples). There were 48 trials in the criterion test: 24 trials involved matched items (e.g., see picture of [kita], hear [kita]) and 24 trials involved mismatched items (e.g., see picture of [kita], hear [qaʃu]), presented in a different random order for each participant. Participants had 3 s to respond YES (the picture and the auditory word are matched) or NO (they are not matched) by pressing keys on the computer keyboard. If participants did not respond within the 3-s time allotment, their answer was considered incorrect, and the program moved to the next trial. The criterion test did not probe participants’ ability to discriminate the velar–uvular contrast, but rather their ability to distinguish completely different words (e.g., [kita]–[qaʃu]); thus the criterion test trials did not involve minimal pairs. Participants were required to reach 90% accuracy on the criterion test before continuing to the final test phase, and they were allowed to complete the word learning–criterion test cycle as many times as necessary to reach the criterion.

Figure 1.

Example criterion test items.

The final test phase trials were identical to those in the criterion test, except that participants were tested on their ability to discriminate the velar–uvular pairs (see Figure 2 for example test items). As in the criterion test, there were 24 matched and 24 mismatched trials. The 24 mismatched trials included 12 trials with [k] auditory forms and [q] pictures and 12 trials with [q] auditory forms and [k] pictures.

Figure 2.

Example final test items.

The entire experiment was conducted in a sound-attenuated booth. Stimuli were presented and responses collected on a computer using DMDX software (Forster & Forster, Reference Forster and Forster2003). Auditory stimuli were presented over headphones, and participants pressed keys on a computer keyboard to register responses.

Results

The mean number of word learning–criterion test cycles required was 3.66 (range = 1–8) for subjects in the Arabic script group and 3.00 (range = 1–6) for subjects in the control group. There was not a significant difference in the number of cycles required to reach criterion between the two groups, t (28) = 1.00, p = .326, indicating that the two groups of subjects did not differ overall in the amount of exposure they had to the new words.

Table 2 presents the results of Experiment 1. An analysis of variance (ANOVA) was conducted with word-learning group (two levels: Arabic script and control) as a between-subjects variable, item type (two levels: matched and mismatched) as a within-subjects variable, and proportion correct as the dependent variable. The main effect of item type was significant, with more accurate performance on matched than mismatched items, F (1, 28) = 135.983, p < .005, partial η² = 0.829. However, neither the main effect of group, F (1, 28) = 0.151, p = .701, partial η² = 0.151, nor the interaction of group and item type, F (1, 28) = 0.024, p = .877, partial η² = 0.001, was significant. In addition, the proportion correct scores were converted to d-primes in order to assess subjects’ ability to detect the difference between matched and mismatched items. These d-primes were submitted to an ANOVA with subject group as a between-subjects variable. Given the proportion of correct results above, the effect of subject group on d-primes was not significant, F (1, 28) = 0.486, p = .492, partial η² = 0.017.

Table 2.

Experiment 1 mean (standard deviation) proportion correct and d-prime scores by word learning group

The results of Experiment 1 thus do not provide evidence that the written forms presented during the word-learning phase impacted subjects’ ability to learn the phonological forms of the novel L2 words. That is, the native English learners did not appear to benefit from the availability of the spelled forms of the words during the word-learning phase when discriminating between the newly learned velar–uvular minimal pairs at test. Given that previous experiments with similar designs and amounts of exposure (e.g., Escudero et al., Reference Escudero, Hayes-Harb and Mitterer2008; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2013) have revealed significant effects of the availability of written forms on memory for the phonological forms of newly learned words, we next sought to understand what might distinguish the present case from the previous studies. To this end, we conducted several follow-up experiments to explore possible explanations for the lack of an effect of written input in Experiment 1. We explored the possibility that the Arabic script was simply uninterpretable to subjects given the combination of its novel symbols and its directionality (Arabic script is written from right to left). In Experiment 2, we attempted to alleviate some of the visual difficulty of the Arabic script by providing explicit instruction about written Arabic prior to the word-learning phase.

EXPERIMENT 2

It is possible that the participants in Experiment 1 did not know where to look when reading the Arabic script because they were unaware that the right sides of the written words held the graphemes that encoded the novel phonological contrast. It is also possible that the participants did not realize that the forms written in the Arabic script contained multiple graphemes, and that they experienced difficulty “extracting” individual graphemes from the written forms. In order to moderate the difficulty of the Arabic script, in Experiment 2, participants were given instruction about the Arabic writing system before they began the word-learning phase.

Procedure

The word-learning, criterion test, and final test phases were identical to Experiment 1. However, before participants began the word-learning phase, they were given instructions about the Arabic writing system. They were told that the Arabic script is read from right to left and not left to right as in English, and completed example trials with an arrow pointing to the right side of the orthographic representation to focus their attention on the right side of the word. Figure 3 presents screen shots from the instruction phase of Experiment 2.

Figure 3.

Unlike English, which is written from left to right, this new language is written from right to left. For example, the two words written in the top panel differ only in their first sounds. The letters representing the first sound appear on the right-hand side of the word. Contrast this with a language such aas English, which is written from left to right, where the letter representing the first sound is on the left side of the word in the second panel. The third panel contains one more example from the new language. The two words differ only in their first letters.

Results

The performance of subjects in this new word-learning condition was compared to that of the control group from Experiment 1 in order to determine whether the addition of explicit instruction about the Arabic script might render the Arabic written forms more helpful to learners. The mean number of word learning–criterion test cycles required for subjects in the Arabic script + instruction condition was 3.250 (range = 1–6). A Mann–Whitney test (given unequal sample sizes) indicated that the difference in number of word-learning cycles required by the Arabic script + instruction group and the control group from Experiment 1 (3.000) was not significant (U = 57, p = .841).

Table 3 presents the proportion correct and the d-prime scores for subjects in the Arabic script + instruction group. A Mann–Whitney test indicated that the difference in d-prime scores between the Arabic script + instruction group (1.026) and the control group from Experiment 1 (1.140) was not significant (U = 58, p = .897).

Table 3.

Experiment 2 mean (standard deviation) proportion correct and d-prime scores

Again, we did not find any benefit of the availability of written forms for this novel contrast and this novel script. Participants in Experiment 2 appear to have performed slightly less accurately than both Arabic script and control subjects from Experiment 1, despite the addition of explicit instruction about the Arabic writing system (though it may be worth noting that Experiment 2 involved only eight participants). In Experiment 3, we attempted to alleviate the script difficulty in a different way: by presenting the written forms in the familiar Roman alphabet.

EXPERIMENT 3

The Arabic velar–uvular contrast may be sufficiently difficult for native English speakers to perceive that even with the availability of (unfamiliar) written forms that support the contrast, native English speakers are unable to learn to distinguish the newly learned minimal pairs. In an effort to isolate the challenge posed by this particular novel phonological contrast, in Experiment 3, we attempted to remove as much script difficulty as possible by presenting written forms using the Roman alphabet. We reasoned that if subjects still have difficulty learning to discriminate the newly learned minimal pairs, it may be that the auditory contrast is so difficult that even when written forms are available, they may simply be unusable.

Results

The performance of subjects in the Roman script condition was compared to that of both the Arabic script group and the control group from Experiment 1 in order to determine whether there are differences in the benefit derived from Roman versus Arabic written forms and whether the availability of Roman written forms improves performance over no written forms. The mean number of word learning–criterion test cycles required for subjects in the Roman script condition was 2.50 (range = 1–5). Mann–Whitney tests reveal that there was no difference in number of word-learning cycles required between the Roman script group and either the Arabic script group (3.667; U = 42.5, p = .245) or the control group (3.000; U = 49, p = .466).

Table 4 presents the proportion correct and the d-prime scores for subjects in the Roman script group. A Mann–Whitney test (given unequal sample sizes) indicated that the Roman script group performed significantly less accurately than both Experiment 1's control group (U = 14, p = .003) and the Arabic script group (U = 22, p = .014).

Table 4.

Experiment 3 mean (standard deviation) proportion correct and d-prime scores

These results suggest that any difficulty associated with the Arabic script was not fully responsible for the lower than expected performance of the learners in Experiments 1 and 2. The availability for written forms in the Roman script appears to have suppressed subjects’ ability to discriminate the newly learned minimal pairs. It is possible that in transliterating Arabic to English in this way, we led subjects to believe that the phonological forms associated with words spelled with <k> and with <q> were not different, because in English both <k> and <q> map to the phoneme /k/ (e.g., “king,” “queen”). In addition, subjects may be familiar with written forms such as “Iraq” (typically pronounced [əɹæk] in American English) and “Qatar” (often pronounced [kɑtɑɹ] or [kʌɾəɹ]). Thus, the written forms may have inadvertently served to confirm their perception of [q] as /k/ in the auditory input.

In Experiment 4, we pursued an entirely different type of modification to the study design: the number of talkers represented in the auditory stimuli. In pilot studies for Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2013), we found that the novel lexical tone contrast was too difficult for subjects to learn when the new words were presented in two voices; however, when we reduced the number of talkers to one (as in the experiments reported in Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2013), subjects were better able to distinguish the newly learned minimal pairs, and differences in performance between subjects in the written forms and the control groups emerged. There are at least two possible explanations for this result: (a) given that the targeted auditory contrast was entirely novel, it may be the case that the variability between the two talkers obscured the variability associated with the novel contrast, and/or (b) one of the two original talkers (i.e., the one removed from the study) produced the lexical tone contrasts in ways that were not as discriminable to subjects, while the talker whose productions were preserved in the ultimate experiments produced them in ways that subjects were able to detect. In the context of the present study, we can explore the effects of two versus one talker, in addition to the individual effects of the two talkers. In Experiment 4, we thus reduced the number of talkers from two to one.

EXPERIMENT 4

Results

The mean number of word learning–criterion test cycles required for subjects in the Talker 1–Arabic script group was 2.400 (range = 1–6) and for the Talker 1–control group was 1.600 (1–3), and the difference between the two groups was significant, t (28) = 2.145, p = .041.

Table 5 presents the proportion correct and d-prime results for these two new word-learning groups. An ANOVA was conducted with word-learning group (two levels: Talker 1–Arabic script and Talker 1–control) as a between-subjects variable, item type as a within-subjects variable, and proportion correct as the dependent variable. The main effect of item type was significant, with more accurate performance on matched than mismatched items, F (1, 28) = 69.803, p < .005, partial η² = 0.714. However, neither the main effect of group, F (1, 28) = 1.464, p = .236, partial η² = 0.050, nor the interaction of group and item type, F (1, 28) = 2.192, p = .150, partial η² = 0.073, was significant. There was similarly no difference between groups in the d-prime scores, F (1, 28) = 0.911, p = .348, partial η² = 0.031. Thus, even when the number of talkers was reduced from two to one, no significant differences in performance emerged between subjects who were presented with written forms and those who were not.

Table 5.

Experiment 4 mean (standard deviation) proportion correct and d-prime scores by word learning group

Talker 1 versus Talker 2

As mentioned above, it is possible that the improved performance of subjects in the Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2013) study over that of subjects in pilot experiments was due to the reduction in the number of talkers from two to one and/or the accessibility of the targeted contrast in the selected talker's speech over that of the other talker. In the interest of determining whether the particular talker chosen impacted subjects’ performance in Experiment 4, we performed one final experiment, which we will call Experiment 4x. Experiment 4x is identical to Experiment 4, except that instead of Talker 1, we used stimuli produced by the second of the two original talkers (Talker 2), in order to determine whether performance on this talker's speech improved over that of Talker 1. In addition, we ran only the Arabic script word-learning condition. Table 6 presents the results of this final experiment.

Table 6.

Experiment 4x mean (standard deviation) proportion correct and d-prime scores by word learning group

The mean number of word learning–criterion test cycles required for subjects in the Talker 2–Arabic script group was 2.533 (range = 1–4), and this did not differ significantly from the Talker 1–Arabic script group, t (28) = –0.325, p = .748. An ANOVA with word-learning group (two levels: Talker 1–Arabic script and Talker 2–Arabic script) as a between-subjects variable and item type as a within-subjects variable revealed a significant effect of item type, F (1, 28) = 104.902, p < .005, partial η² = 0.789, a nonsignificant (but marginal) effect of word-learning condition, F (1, 28) = 3.963, p = .056, partial η² = 0.124, and a significant interaction of the two, F (1, 28) = 7.738, p = .010, partial η² = 0.217, on proportion correct scores. However, there was no significant difference in d-prime scores between the two word learning-groups, F (1, 28) = 2.859, p = .102, partial η² = 0.093. Together, we may interpret these findings as indicating that while subjects may have performed slightly more accurately on Talker 1's than Talker 2's voice with respect to mismatched items, the Talker 1 and Talker 2 word-learning groups did not differ overall in their ability to discriminate the newly learned minimal pairs.