PROCESSING INSTRUCTION

PI is a pedagogical intervention that is informed by the strategies L2 learners use to process input, specifically, VanPatten’s model of input processing (VanPatten, Reference VanPatten1996, Reference VanPatten and VanPatten2004, Reference VanPatten, VanPatten and Williams2015a). This model makes predictions about what L2 learners do and do not do as they attempt to process sentence-level input. PI is an input-oriented intervention: Learners do not produce target items during training. Instead, they are given activities called SI to assist them with correct interpretation of sentence structure and to make correct form-meaning connections. The SI activities are designed so that learners must rely on sentence structure or a target form to correctly interpret the meaning that the form or structure encodes (VanPatten, Reference VanPatten2015b; Wong, Reference Wong and VanPatten2004a). In its original conception, PI also included explicit explanation about the targeted structure as well as a warning about the incorrect processing strategies that learners typically use that could lead to incorrect processing of target forms. For example, one of the predictions of the model is the First-Noun Principle (FNP): Learners tend to process the first noun or pronoun they encounter in a sentence as the agent (VanPatten, Reference VanPatten, VanPatten and Williams2015a). The tendency to assign the role of agent to the first animate noun phrase is known to be crosslinguistically common in the comprehension of structures with noncanonical word orders (e.g., passive forms, scrambled sentences, object clefts, and object-relative clauses), and the effect of this tendency has been demonstrated in children as well as in individuals with specific language impairment (SLI) or aphasia (Grodzinsky, Reference Grodzinsky2000; Leonard, Wong, Deevy, Stokes, & Fletcher, Reference Leonard, Wong, Deevy, Stokes and Fletcher2006; Swinney & Zurif, Reference Swinney and Zurif1995; cf., Meyer, Mack, & Thompson, Reference Meyer, Mack and Thompson2012). This strategy is effective when learners are processing SVO order sentences but will lead to incorrect interpretations if they are processing sentences in which the first noun is not the agent. This is the case with the French causative construction.

1. (1) Pierre fait faire du café à Marie.

   Pierre makes to make coffee to Marie.

   “Pierre makes Marie make coffee.”

2. (2) Jean fait acheter du lait à Paul.

   Jean makes to buy milk to Paul.

   “Jean makes Pierre buy milk.”

The verb faire is a common verb in French that means to do or to make (e.g., Pierre fait du café/Pierre makes coffee). This verb is also used as the main verb in the causative construction. As illustrated in (1) and (2), the causative construction consists of a conjugated form of the verb faire as the main verb followed by a second verb in the infinitive form to specify the caused action. In (1) the agent of the first conjugated verb, fait, is Pierre. The agent of the second verb in the infinitive form, faire, is Marie. This second agent Marie is placed postverbally and marked by the preposition à. Thus, the agent of the action faire du café is Marie. In (2) the agent of the main verb fait is Jean while the agent of the second infinitive verb infinitive, acheter, is Paul. The agent of the action acheter du lait is thus Paul. As verified in VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004), when L2 learners process sentences like (1) and (2), they tend to interpret Pierre as the agent of the act of making coffee and Jean as the agent of buying milk because Pierre and Jean appear in the canonical subject position and are good candidates for executing the actions. VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004) attempted to help L2 French learners overcome this processing strategy by first giving them EI about the causative structure and by making them aware of this tendency, which would lead to incorrect sentence interpretation for the French causative. Following this instruction, learners were given SI activities to help them overcome this tendency. These activities required them to listen to and read sentences that contained the verb faire in the causative and noncausative construction and then to determine who was performing the action. Participants never produce the target structure in these SI activities. This type of SI practice was designed so that learners would no longer strictly rely on the first noun as the agent of the action. Because cues such as animacy and plausibility were absent from sentences, the only way learners could correctly identify the agent of the caused action was through processing sentence structure.

The cue that learners needed to pick up to interpret that a sentence was causative was the presence of the second verb in the infinitive form after the conjugated main verb fait because it is this second verb that distinguishes causative sentences from noncausative sentences. For example, in (1), when learners initially only hear Pierre fait..., the sentence could plausibly be interpreted as Pierre doing something (e.g., Pierre fait du café [Pierre makes coffee]) or that he is going to have someone else do something (e.g., Pierre fait faire…). However, if learners subsequently hear a second verb in the infinitive (in this case faire), the only possible correct interpretation is that Pierre is going to have someone else do something. The goal of SI activities for the causative structure is to train learners to pick up on the cue that “fait + infinitive verb” denotes a causative construction that signals that the first noun cannot be the agent of the action. If such training is effective, learners who initially rely on the FNP should revise their sentence interpretation when they come across a second verb in the infinitive after the main verb fait.

In VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004), PI training was compared to a drill-based treatment that they called TI, which consisted of giving participants EI on the causative structure followed by production practice with mechanical, meaningful, and communicative drills (see Paulston, Reference Paulston, Allen and Campbell1972 for rationale behind this practice). For example, participants in the TI group would see Marc/faire les valises à Jean (Marc/to pack the bags to Jean) and would have to create the causative construction with the elements: Marc fait faire les valises à Jean (Marc makes Jean pack the bags).

Using a pretest-posttest design, participants were assigned to either a PI, TI, or a control group. Training items for both treatment groups were balanced for vocabulary, number of activities, and practice time. The difference was that while the PI group received practice in interpreting sentence structure with SI activities, the TI group’s practice involved producing the target structure through drill practice. Regarding the EI that the two groups received, the difference was that the PI group additionally received a description of the difficulty that French learners have with this structure due to their FNP tendency. VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004) explained that the TI group was not given this information because this was not typically part of traditional approaches to grammar instruction (p. 103).

The comparison of interpretation accuracy before and after training revealed the PI group performed significantly better than the TI group, while the two groups performed similarly on a sentence production task. Other studies that have found similar advantages of PI over TI include VanPatten and Cadierno (Reference VanPatten and Cadierno1993), Cadierno (Reference Cadierno1995), Benati (Reference Benati2001), and Comer and deBenedette (Reference Comer and DeBenedette2010). Interestingly, VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004) also found that many TI participants also made the noncausative sentences causative, that is, they overgeneralized the newly learned structure and applied it to all items, even to distractor items that did not take the causative structure (see VanPatten & Wong, Reference VanPatten, Wong and VanPatten2004 for details). The PI group did not do this. When those TI participants who made distractor items causative were removed from the analysis, the production performance of the PI group (who had no practice in production) was better than the TI group, which performed no better than the control group. VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004) concluded that TI teaches learners a test-taking strategy while PI teaches correct input processing that may lead to changes in mental representation enabling subjects to produce correct structures even though they did not practice doing so during training.

EFFECT OF EXPLICIT INFORMATION

A criticism of earlier research comparing PI with TI was that the group factor was confounded with instruction manipulation: The PI group was explicitly told about the FNP strategy while the TI group was not. VanPatten and Oikkenon (Reference VanPatten and Oikkenon1996) pointed out that the superior results found for PI in VanPatten and Cadierno (Reference VanPatten and Cadierno1993) could be due to the type of EI they received and not to the SI practice. Therefore, VanPatten and Oikkenon (Reference VanPatten and Oikkenon1996) isolated EI as a variable in PI. In their study on Spanish object pronouns with OVS order, they compared three groups: (1) EI + SI, (2) SI only without EI, and (3) a group that received EI only with no activities. They found that the group that received EI and SI, and the SI only group that did not get EI, performed the same. The EI only group made no gains. These results led the researchers to conclude that EI played no role: The component of PI that is responsible for the positive effects of PI research are the SI activities and not the EI. Many following studies have since replicated the results of VanPatten and Oikkenon (e.g., Benati, Reference Benati and VanPatten2004; Sanz, Reference Sanz and VanPatten2004; Sanz & Morgan-Short, Reference Sanz and Morgan-Short2004; Wong, Reference Wong and VanPatten2004b).

Fernandez (Reference Fernandez2008) introduced a different dimension to research on the role of EI in PI by examining whether EI can speed up the correct processing of certain structures. Using trials-to-criterion, a measure that counts how many items it takes for learners to begin processing a structure correctly, Fernandez compared PI with and without EI on two structures in Spanish: the subjunctive and object pronouns with OVS order. The results revealed that for the OVS structure, EI played no role in how soon, fast, or accurately participants processed the structure, supporting VanPatten and Oikkenon’s findings (1996). For the subjunctive, however, results showed that participants who had EI began to process the structure sooner, faster, and more accurately than those who did not have EI. Participants who only had SI did also eventually begin to process the subjunctive sentences correctly, confirming that SI practice alone can induce correct processing. Because having access to EI benefited the learners with the subjunctive structure but not for the OVS structure, Fernandez (Reference Fernandez2008) concluded that the benefits of EI may depend on the nature of the target structure.

VanPatten, Borst, Collopy, Qualin, and Price (Reference VanPatten, Borst, Collopy, Qualin and Price2013) also set out to isolate the effects of EI in PI as one of the goals in their study involving Spanish, German, French, and Russian. In a series of four experiments, they compared participants who received both EI and SI to participants who only received SI. Each language addressed a different target structure but all four structures shared the same processing problem, namely the FNP. The dependent measures were trials-to-criterion and improvement (i.e., posttest minus pretest) scores. In the Spanish study involving object pronouns with OVS order, there was no difference on either measure between participants who received EI with SI, and participants who received SI only without EI, replicating the results of Fernandez (Reference Fernandez2008). In the German study with nominative/accusative case marking on definite articles, participants who received EI before engaging in SI practice began to process sentences correctly sooner than those who did not have access to EI. On the pretest-posttest analysis, there was a marginal difference between the +EI and the –EI groups. This suggests that the –EI participants may have started to catch up to the +EI group but perhaps needed more items to do so (VanPatten et al., Reference VanPatten, Borst, Collopy, Qualin and Price2013, p. 516) (see also Culman, Henry, & VanPatten, Reference Culman, Henry and VanPatten2009, and Henry, Culman, & VanPatten, Reference Henry, Culman and VanPatten2009).

In the Russian experiment targeting nominative/accusative case marking on nouns, EI did not play a significant role in either trials-to-criterion or in the outcome based on the pretest-posttest measure. The French experiment, however, that targeted the causative structure as in VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004), showed that the EI tended to help subjects process the structure sooner than those in the –EI condition. No significant difference, however, was found between the two conditions in the pretest-posttest analysis. This suggests that although EI might have helped participants process the target structure sooner, EI was not a factor in the outcome because it appears the –EI group was catching up to the +EI group in the end. The overall conclusion of this study across all four languages is that EI can speed up correct input processing for some structures (i.e., the French causative and German case marking). At the same time, because all participants in the –EI conditions eventually processed the structures correctly, EI may not be critical. SI alone appears to be sufficient to induce correct processing.

Henry (Reference Henry2017) examined the effects of EI and prosody on the same German target structure as VanPatten et al. (Reference VanPatten, Borst, Collopy, Qualin and Price2013) and found that EI helped learners process correctly sooner. Additionally, Henry (Reference Henry2017) found that participants who received either EI or prosody cues improved on an immediate comprehension and production task and this improvement was maintained on a delayed comprehension task four weeks later. The –EI participants, however, only improved on the immediate comprehension posttest and this improvement was not maintained on delayed posttests. Because Henry (Reference Henry2017) included delayed posttests, this study has evidence to show that EI can help learners process this target structure correctly sooner and maintain correct processing of the structure over time.

EYE-TRACKING: AN ONLINE MEASURE OF SENTENCE PROCESSING

In psycholinguistic research, eye-tracking has long been used as an online measure to assess how people process spoken input. The present study makes use of a visual world paradigm (see Trueswell & Tanhenhaus, Reference Trueswell and Tanenhaus2005, for a summary of its history) where participants’ gazes on visual targets (such as objects laid out or projected on the screen in front of them) are traced while they listen to speech input. The visual world paradigm has been particularly useful for revealing incremental online syntactic processing and has demonstrated that listeners integrate the grammatical role and the meaning of an incoming word as soon as sufficient sound information of the word unfolds in speech (Altmann & Kamide, Reference Altmann and Kamide1999; Kamide, Altmann, & Haywood, Reference Kamide, Altmann and Haywood2003). For the purpose of the current study, we adopted the paradigm that has been used to test online comprehension of passive constructions in typically developing children and adults (Stromswold, Eisenband, Norland, & Ratzan, Reference Stromswold, Eisenband, Norland and Ratzan2002) and individuals with aphasia (Meyer et al., Reference Meyer, Mack and Thompson2012), where two nearly identical scenes that differ only in the thematic roles of two people were presented side by side while participants listen to the sentence (e.g., a picture of a boy shaving the face of a man and a picture of a man shaving the face of a body are presented simultaneously with the target sentence “The man was shaved by the boy”). Participants were simply asked to choose the picture that matched the sentence they heard.

Using this picture-match paradigm, Stromswold et al. (Reference Stromswold, Eisenband, Norland and Ratzan2002) showed that adults swiftly shifted their gaze to the correct picture as soon as they heard the main verb of the passive form (e.g., The girl was pushed) while children did not look at the correct picture until they heard the prepositional phrase that specified the agent (by the boy). In Meyer et al. (Reference Meyer, Mack and Thompson2012), participants with Broca’s aphasia did not show the looks to the incorrect picture where the subject is the agent (e.g., A man is shaving a boy) while listening to a passive sentence (e.g., The man was shaved by a boy). In contrast, their chronological-age-matched control participants showed a clear bias toward the “first noun = agent” pictures until the end of the main verb (shaved). In the present study, L2 learners of French viewed two scenes while they listened to the causative structure (Pierre fait faire du café à Marie. Pierre makes Marie make coffee), which tends to be incorrectly interpreted with the FNP (VanPatten & Wong, Reference VanPatten, Wong and VanPatten2004).

Given the repeatedly observed FNP bias, we predict initial incorrect looks to the scene where the first noun plays the agent role (e.g., “Pierre…” will send listeners’ eyes to the picture where the man is making coffee for the woman) before learners receive any kind of intervention. The primary question is whether different types of interventions (PI or TI) lead to differences in online processing of the causative structure. Aforementioned previous research suggests that participants who are trained with PI are more successful on sentence interpretation tasks than participants who are trained with TI (Benati, Reference Benati2001; Cadierno, Reference Cadierno1995; VanPatten & Cadierno, Reference VanPatten and Cadierno1993; VanPatten & Wong, Reference VanPatten, Wong and VanPatten2004). Thus, we predict that those who receive SI activities may show reductions in initial looks to the incorrect picture (due to a faster switch to the correct picture) to a larger degree than those who receive drill-based TI activities.

The second question concerns the effect of EI on the change in online processing. While research has shown that the EI in PI is not necessary for correct input processing (Benati, Reference Benati and VanPatten2004; Sanz & Morgan-Short, Reference Sanz and Morgan-Short2004; VanPatten & Oikkenon, Reference VanPatten and Oikkenon1996; Wong, Reference Wong and VanPatten2004), it may help learners correctly process some structures more quickly (e.g., Henry, Reference Henry2017; VanPatten et al., Reference VanPatten, Borst, Collopy, Qualin and Price2013). The present study compared changes in fixation patterns across two groups that did not receive EI (Experiment 1, –EI) and who did receive EI (Experiment 2, +EI). If EI facilitates online processing of the causative structure in French regardless of the type of intervention, the reduction of initial incorrect fixations after training should be larger for both PI and TI groups in Experiment 2 than in Experiment 1.

RESEARCH QUESTIONS

METHOD

MATERIALS

EYE-TRACKING STIMULI

Ninety-six pairs of pictures were created using the Bitstrips application (https://www.facebook.com/Bitstrips/), which provides templates for creating and designing comics. The pictures always depicted the same two people, a male (Pierre) and a female (Marie). Each item had a pair of identical scenes across which the actions/positions of the two people were swapped (Figure 1).

FIGURE 1.

Sample slide for eye-tracking stimuli.

Twenty-four paired pictures were presented as target items in the causative and the noncausative conditions. Each of these critical items was presented with a set of two causative sentences (e.g., Pierre/Marie fait faire du café à Marie/Pierre. “Pierre/Marie makes Marie/Pierre make coffee.”) and two noncausative sentences (Pierre/Marie fait des café pour Marie/Pierre. “Pierre/Marie makes coffee for Marie/Pierre.”) across four lists such that each participant hears only one of the four sentences that can be matched with one side of the picture pair. An additional 72 items were distractors. Of these distractors, 24 pairs of pictures were presented across the four lists with two active sentences (Pierre/Marie tape Pierre/Marie. “Pierre/Marie pokes Marie/Pierre.”) and two passive sentences (Pierre/Marie est tape(e) par Marie/Pierre. “Pierre/Marie is poked by Marie/Pierre.”).^{Footnote 1} Forty-eight other distractors contrasted the actions of the two people in the scene (e.g., Pierre/Marie dort, et Marie/Pierre danse. “Pierre/Marie sleeps, Marie/Pierre dances.”). One third of these items were combined with sentences with an object relative clause (e.g., Pierre/Marie sent le t-shirt que Marie/Pierre a lavé. “Pierre/Marie smells the t-shirt that Marie/Pierre washed.”).

A total of 288 sentences (48 causative, 48 noncausative, 48 active, 48 passive, 64 conjoined, and 32 object-RC) were recorded by a female native speaker of French from Rennes, France, at 44.1 KHz sampling rate using Praat (Boersma & Weenink, 1992–2014).

PI training

A total of 12 sentences containing the verb faire in the causative construction and 12 sentences with the verb faire in the noncausative construction were modeled after the SI activities in VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004). Each sentence was presented on a slide with an English comprehension question and two possible answers (Figure 2).

FIGURE 2.

Sample slide for PI training.

Participants responded by pressing a button on a button box. Once either a or b was chosen, the display showed the correct answer in red to provide feedback. No other information about the structure was given. Participants pressed the rightmost button to move on to the next sentence. Participants in the PI group never produced the target structure.

TI training

A total of 24 items were modeled after the TI treatment materials in VanPatten and Wong (Reference VanPatten, Wong and VanPatten2004). Participants were asked to create causative sentences with the verb faire following the model by typing their sentences into a box (Figure 3). Each prompt appeared on the screen individually. When participants finished typing their answer, they pressed the space bar to see the correct answer. No other information about the structure was given. Participants pressed the space bar to move on to the next prompt.

FIGURE 3.

Sample slide for TI training.

Training items in both groups were balanced for vocabulary and number of activities. The difference was that the PI group received SI practice in interpreting sentence structure while the TI group created causative sentences by following a model (as is typical with traditional drill practice).

ANALYSES AND RESULTS

Data from two participants were excluded from the analysis due to incompletion of the session and a system error. The following reports data from 32 participants in the TI group and 30 participants in the PI group.

Accuracy

The effects of block and training type on mouse click responses were tested with a mixed-effects logistic regression model, which specified the training type (TI or PI) and the block (before or after training) and their interaction as predictor factors. Accuracy of picture selection for the causative structure sentences was very low for both TI and PI groups before training with no significant difference between the two groups (TI 3.9%; PI: 4.6%: Est = –.12, SE = .68, z = –.18, p = .86: Figure 4). After training, accuracy in the PI group improved to 62% while the TI group only improved to 11% (Figure 5). The result showed a robust effect of block (z = 7.25, p < .001), of training type (z = 5.07, p < .001) and a significant interaction between the two (z = 5.26, p < .001). The effect of block was significant for both the TI (z = 4.03, p < .001) and the PI training (z = 8.18, p < .001).

FIGURE 4.

Experiment 1: Accuracy of picture selection before training.

FIGURE 5.

Experiment 1: Accuracy of picture selection after training.

Eye-Movement Patterns

Because picture selection accuracy showed a robust effect for training type, the eye movements that preceded the mouse clicks are expected to reflect the effect of training as well. The important question is whether the fixation patterns will capture the expected initial bias toward the picture that would be congruent with the “first noun = agent” interpretation, and its predicted reduction after training.

Because participants were viewing either the correct or the incorrect picture at any point in time, the changes in gazes were expressed in logit, or the log of the ratio between the numbers of fixations on the two areas of interest. The logit functions were plotted with the baseline of 0 in the middle of y-axis such that the relatively higher numbers of fixations on the correct picture would be shown above the midline whereas the relatively higher number of incorrect fixations would be shown below the midline. Thus, the stronger the FNP bias is, the larger downward deviation of the function from the midline we should observe. The fixation logits were aligned at the utterance onset, and at the onsets of critical windows according to the duration information of each auditory stimulus.

Figure 6 shows the TI and PI groups’ fixation patterns while they listened to the target causative structure in the before-training block (upper panels) and in the after-training block (lower panels). The two upper panels show the gradual increase in incorrect fixations to the “first noun = agent” picture throughout the causative sentence in both groups. The two lower panels for the after-training blocks, however, show a contrast between the training types: While the TI group showed a gradual increase in looks to the incorrect picture to a slightly greater degree than in the before-training block, the PI group showed a decrease in the incorrect fixations toward the end of the sentences. Importantly, the turning point in the fixation logits appeared at the beginning of the prepositional phrase (à Marie/Pierre), that is, participants in the PI group have started to shift their gazes before they processed the entire sentence.^{Footnote 2}

FIGURE 6.

Experiment 1: TI and PI groups’ fixation patterns while listening to the causative structure in before-training block (upper panels) and after-training block (lower panels).

For the statistical analysis, the empirical logit (elogit) (Barr, 2008) was calculated for each time window (w1: subject Pierre/Marie; w2: fait; w3: faire; w4: the rest of the verb phrase; w5: à Marie/Pierre) of each trial. A set of mixed-effects models tested the effects of training type, block, and the interaction between the two on the elogits for each window (the training type and block were both sum coded). The p-value of each effect was tested through comparison of the models that were minimally different with the presence or absence of the factor in question. Table 1 summarizes the results for each time window.

TABLE 1.

Experiment 1: Summary of mixed-effect regressions on the fixation logits Elogit ∼ Training + Block + Training:Block + (1+Block|subj) + (1+Training|item)

N = 712: total number of observations participated in the model. ***p < .001, **p < .01, *p < .05, ^p < .1.

The constantly significant intercept across the five windows indicates an overall bias toward the incorrect picture throughout the sentences regardless of the training type and block. For the last window (à Marie/Pierre), a significant effect of training and a marginal interaction between training type and block were confirmed, indicating that the PI group made relatively more fixations to the correct pictures than the TI group, and the PI group made a larger increase in the looks toward the correct pictures across blocks than the TI group. The marginal effect of block in the last window was found in the PI group (z = 1.91, p = .06) but the TI group showed no effect of block (z = 1.21, p = .27).

DISCUSSION

Evidence for the FNP bias was confirmed both in eye-movement patterns and the following picture selection: Before training, participants increasingly gazed at the incorrect “first-noun = agent” picture throughout the causative sentence, leading to less than 5% accurate picture selection (i.e., the answer to RQ1 is yes).

The results of Experiment 1 also confirmed the previously reported effect of training type (VanPatten & Wong, Reference Wong and VanPatten2004): The SI activities in the PI training that directed participants’ attention to who did what in each sentence significantly improved the processing and comprehension of the target causative structure as compared to the TI training that forced mechanical production of causative sentences. The effect of PI on the accuracy of picture selection was robust even without explicit instruction on the target structure.

Importantly, the FNP bias in eye movement was reduced after training for the PI participants only. A closer examination of the gaze patterns further suggests that while the PI activities may have changed participants’ processing mechanism for the target structure, the TI activities did not lead to any changes in processing for the TI participants. Although accuracy of picture selection showed a minute improvement in the TI group, there was no visible change in their eye-fixation patterns after training. In contrast, the PI group showed the turning point in the fixation logit function at the onset of the final prepositional phrase (à Marie/Pierre).

The finding that PI participants started to shift their gaze to the correct picture at the onset of the final prepositional phrase à Marie/Pierre could lead one to speculate if participants might have been relying more on “à + noun” as a cue rather than “fait + infinitive verb.” Besides the second infinitive verb, another cue that a sentence is causative is the presence of “à + noun.” Compare noncausative sentence (3) to causative sentence (4):

1. (3) Marie fait la vaisselle avec Pierre.

   Marie does the dishes with Pierre.

2. (4) Marie fait faire la vaisselle à Pierre.

   Marie makes to do the dishes to Pierre.

   “Marie makes Pierre do the dishes.”

However, because it takes 150 to 200 ms to program and execute the saccades (see Trueswell & Tanenhaus, 2005, for a review), the observed timing of the gaze shift suggests that PI participants started entertaining the opposite interpretation based on the information unfolding before the final prepositional phrase (à Marie/Pierre) that may have more clearly disambiguated the sentential meaning. Therefore, taken together, the present experiment has shown the effect of PI in both the online and offline measures (RQ2): PI practice appeared to change the way that the PI group processed the target structure but the TI practice did not lead to any processing changes for the TI group.

RESEARCH QUESTIONS

METHOD

The same method used in Experiment 1 was used for Experiment 2. However, in Experiment 2, a slide that explained the target structure was added at the beginning of the TI and the PI training (Figure 7).

FIGURE 7.

EI slide for Experiment 2.

ANALYSES AND RESULTS

Data from one participant was removed from the analysis due to a file coding error. The following reports results from 25 participants in the TI group and 24 participants in the PI group.

Accuracy

Results for accuracy were similar to Experiment 1: Both TI and the PI groups had low accuracy in picture selection before training, but the PI group made a larger increase in accuracy after training than the TI group (Figures 8 and 9). The mixed effects logistic regression confirmed the lack of group difference in the before-training block (z = .007, p = .995), a main effect of training type (z = 2.78, p < .01), a main effect of block (z = 9.07, p < .001), and an interaction between the two (z = 2.91, p < .01).

FIGURE 8.

Experiment 2: Accuracy of picture selection before training.

FIGURE 9.

Experiment 2: Accuracy of picture selection after training.

Compared to Experiment 1, both groups exhibited larger increases in picture selection accuracy after training. The PI group reached 77% accuracy (as compared to 61% in Experiment 1), and even the TI group showed 31% accuracy (Experiment 1: 11%). Thus, the EI appeared to boost accuracy in both groups. The mixed-effect logistic regression on the data from both experiments showed a main effect for training (z = 4.24, p < .0001), a main effect for block (z = 11.13, p < .0001), and an interaction between the two (z = 5.06, p < .0001). The factor experiment (+EI/–EI), however, did not show a significant main effect (z = 1.2, p = .23), and had a marginal interaction with block (z = 1.96, p = .05). Within each training type (tested in separate within-group models), there was no interaction between block and experiment for either the TI group (z = .91, p = .36) or the PI group (z = .94, p = .35). Therefore, the improvement of accuracy after training was not significantly larger for Experiment 2 (+EI) as compared to Experiment 1 (–EI).

Eye-Movement Patterns

The fixation logit functions obtained from Experiment 2 confirmed a general strong bias toward the “first noun = agent” picture like in Experiment 1. Participants in both groups gradually shifted their gazes to the incorrect picture throughout the causative sentences during the block before training (upper panels of Figure 10).

FIGURE 10.

Experiment 2: TI and PI groups’ fixation patterns while listening to the causative structure in before-training block (upper panels) and after-training block (lower panels).

The outcome of Experiment 2 differed from Experiment 1 in the gaze patterns after training. Unlike Experiment 1, participants in both TI and PI groups showed a reduction in looks to the incorrect “first noun = agent” toward the end of the causative sentences (Figure 10 lower panels). Importantly, in both groups, the logit function showed the turning point from which the looks to the incorrect picture gradually reduced at the beginning of the prepositional phrase (à Marie/Pierre), which was shown only for the PI group in Experiment 1. The number of incorrect fixations seemed relatively smaller in the PI group than in the TI group (notice that the looks to the two pictures reached the 50-50 level in the PI group by the end of the target sentences). The mixed-effects models revealed a main effect for block and a marginal effect for training type, and no interaction between the two for the last analysis window (Table 2). Within each group, the effect of block was not significant (TI: z = 1.01, p = .3; PI: z = 1.2, p = .19), indicating that the size of the reductions in the biased looks was relatively small.

TABLE 2.

Experiment 2: Summary of mixed-effect regressions on the fixation logits Elogit ∼ Training + Block + Training:Block + (1+Block|subj) + (1+Training|item)

N = 587: total number of observations participated in the model. ***p < .001, **p < .01, *p < .05, ^p < .1.

To test the effect of EI, another set of mixed-effects models was fitted to the data from both experiments. The results showed a robust main effect for training type and a marginal effect for experiment (i.e., +EI/–EI) for the last window (à Marie/Pierre), indicating that the PI groups in both experiments had a significantly smaller number of fixations to the incorrect picture before the end of the sentence, and there were overall fewer instances of incorrect looks to the “first-noun = agent” picture in the last window when participants received EI (Table 3). When the effect of block and experiment (+EI/–EI) were tested within each training type, the TI group showed neither the effect of block (z = .02) nor the effect of experiment (i.e., EI) (z = .77), nor the interaction between the two (z = 1.52, p = .13). The PI group showed a marginal effect of block (z = 1.9, p = .05), no effect of EI (z = 1.32, p > .1), and no interaction between the two (z = 0.03, p > .1).

TABLE 3.

Experiment 1 and 2: Summary of mixed-effect regressions on the fixation logits Elogit ∼ Training + Block + Experiment + Training*Block*Experiment + (1+Block|subj) + (1+Training|item)

N = 1282: total number of observations participated in the model. ***p < .001, **p < .01, *p < .05, ^p < .1.

DISCUSSION

The results of Experiment 2 confirmed the general FNP tendency that was shown in Experiment 1 in both the picture selection task and eye-movements patterns in the before-training blocks (RQ1). The similarities in performances in the before-training blocks across the two experiments confirm that the level of spoken sentence comprehension and the initial FNP parsing strategy of the L2 learners of French were comparable across the two experiments.

The effect of training type to attenuate the FNP tendency was confirmed in the picture selection task (RQ2): The PI + EI group exhibited a larger improvement in accuracy scores than the TI + EI group. In the eye-tracking data, both the TI and PI groups showed a reduction in looks to the incorrect (“first noun = agent”) pictures toward the end of the causative sentences after treatment, which was only observed in the PI group in Experiment 1. This suggests that the EI in Experiment 2 may have helped TI participants reduce the amount of gazes to the biased “first noun = agent” picture. While the number of incorrect fixations seemed smaller in the PI group than in the TI group, this difference was not statistically significant.

As for the effect of EI, the group difference in accuracy improvement seemed much larger in Experiment 2 than in Experiment 1. However, the statistical analysis on the data pooled across the two experiments revealed only a marginal interaction between block and experiment, and not a three-way interaction between training type, block, and experiment, which would have been shown if EI had enhanced the effect training more robustly for the PI group. Therefore, EI seemed to have little benefit for both the PI and the TI groups in terms of accurate picture selection.

The lack of effect of EI on online processing was also confirmed in the eye-tracking data of Experiment 2. After participants’ attention was drawn to the target structure, the looks to the “first noun = agent” picture showed the reduction or the turning point before the end of the sentence regardless of training type. Importantly, the timing of the turning point was similar across groups – participants started entertaining the other interpretation from the beginning of the last phrase. Hypothetically, in the present study, the EI combined with the PI training could have led to faster processing of the target structure. If the PI-based training with EI had immediately modulated the parsing efficacy and enhanced the certainty for the causative interpretation, participants’ incorrect interpretation could have been revised immediately after they heard the structural cue—the verb sequence “fait faire,” and the looks could have been more clearly directed toward the correct picture (i.e., the function could have flipped to show a rise above the midline). However, the data did not confirm such an additive or interactively enhanced effect. The present results therefore confirm the resilience of the FNP bias in online processing of the causative structure in L2 French learners.