Keywords

1 Introduction

Our study was aimed at defining strategies of paired associate vocabulary learning. Second language vocabulary learning is a part of second language acquisition (SLA). SLA refers both to the study of individuals and groups who are learning a language subsequent to learning their first one (L1) as young children, and to the process of learning that language. The additional language is called a second language (L2), even though it may actually be the third, fourth, or tenth to be acquired [24: 2]. The studies on SLA demonstrated that learners do apply different learning strategies [1, 16]. But learning strategies are not to be understood as personal styles of learning. We understand strategies as a particular configuration of cognitive processes and mechanisms, which enable problem-solving and which change under certain conditions.

1.1 The Studies of Bilingual Lexicon

Cognitive theories seek to explain SLA in terms of mental representations and information processing. In contemporary research the problem of second language (SL) vocabulary learning is considered as bilingualFootnote 1 lexicon formation, which is viewed as mental “storage” of word forms and meanings. Research in the field of the bilingual mental lexicon began in the 1980s, when a number of models appeared describing the organization and links of lexical units of different languages. According to the word-association model, for instance, there is a direct connection between L1 and L2 lexical units—words are connected as translation equivalents [11]; other models, such as the concept-mediation model, postulated that these connections are meaning-mediated [20]. Contemporary research on the bilingual mental lexicon is flourishing (for reviews and textbooks see: [17, 19]).

Cognitive psychology and psycholinguistic research address two aspects of the bilingual mental lexicon problem: word recognition and semantic level access [4], and mental lexicon formation and organization [6, 9]. The research on mental lexicon formation mostly deals with defining the role of two groups of factors in the vocabulary acquisition process: external factors, such as presentation format, context and instruction [12]; and the use of learning strategies [1, 9, 16].

Most of methods of the bilingual lexicon investigation are based on the analysis of the lexicon organization via performance indicators (see [19]), such as reaction and execution time, percentage of correct answers. The tasks used here include lexical decision, semantic priming, picture naming, semantic categorization tasks and variations of the Stroop test. Apart from the experimental procedures there is another group of methods, which simulate ecologically valid tasks, such as narrative methods, reading and translation. In recent research these tasks have been accompanied by the recording of objective characteristics of visual material processing as it occurs, such as eye tracking (see [23]). In our study the eye tracking technology will be employed to tackle the problem of bilingual lexicon formation.

1.2 Eye-Tracking Technologies in the Studies of Bilingual Lexicon

Eye-tracking has a number of advantages compared to traditional research methods of the bilingual lexicon: first, registration of eye movements can be used in analysis of natural tasks, such as reading and visual word recognition; secondly, recognition time and other performance indicators of text comprehension can be assessed through their indirect manifestations; finally, eye movement registration makes it possible to obtain objective measures of cognitive resources distribution that can be associated with the subjects’ free report [22].

In the majority of eye-tracking research on SL acquisition, a reading technique has been used. The most common eye-movement measures used in a variety of tasks that include SL acquisition and processing are (see [7, 21, 23]).

Eye tracking technique has been broadly used in bilingual studies in tackling such issues as reading SL texts, lexical representation in bilinguals, grammatical and discourse processing. Some of the issues in bilingual lexicon studies that benefit from eye movement recording are:

  • Processing difficulties in reading the SL context [21, 22];

  • Lexical representation and access in bilinguals: semantic constraint effect in homograph recognition (words with similar forms, but different meanings in L1 and L2) and cognate facilitation (words with overlapping orthographical and semantic representations in L1 and L2) [13, 25, 26];

  • Research on grammatical and discourse processing revealed, for instance, that bilinguals perform a complete syntactic parsing of sentences when reading in the second language, and they do so in a manner similar to native speakers [5].

One of the possible applications of bilingual lexicon studies is a project, iDICT, developed by Finnish computer scientists in collaboration with SensoMotoric Instruments [8]. The intellectual interface is designed to detect eye movement characteristics associated with processing difficulties (regressions, fixation duration) and to provide translation when needed.

2 Experiment

In our study, we consider a relatively simple procedure of foreign vocabulary learning: what takes place when new words are presented in pairs with their translation into the subjects’ native language.

In this situation, a foreign word must be linked to the meaning given by the word of the native language, in order to be recalled later. Several considerations influenced our focus on the study of this particular learning task. Firstly, it is the main source of vocabulary learning at the initial stages of second language acquisition in adults (in bilingual second language acquisition children learn new words mostly from the context). The basis of the bilingual lexicon in adults is formed by the non-contextual method, which speeds up language acquisition. Secondly, this method allows discovering the basic connection patterns of the new word form with its meaning. Thirdly, paired associate vocabulary learning has been studied much less than contextual vocabulary acquisition.

Primary research on paired associate learning, with the use of eye tracking technique, was carried out by McCormack et al. [14]. However, there have been very few subsequent studies of paired associate learning with eye movements recording since then (especially compared to contextual vocabulary acquisition and recognition). Our work is intended to (a) fill the gap in the eye movement research of vocabulary learning and of paired associate learning in general, and (b) reveal with the help of eye tracking technologies, cognitive strategies used in paired associate vocabulary learning.

We used the paired associate technique introduced by Calkins [2]: the subjects were presented with pairs of stimuli, and then, based on the test stimulus (one of the pair), they had to recall the second stimulus. In our study the stimuli were pseudo-words and their “translation” into the native language of the subjects (Russian). We suggest that the effectiveness of memorizing new words is linked to (a) the mnemonic method used, (b) the order of word presentation, and (c) eye-movement characteristics.

2.1 Participants and Experimental Setup

Subjects. 31 university students, aged 18–26, experienced English learners (B1–C1 level of English), 23 females, 8 males.

Stimuli. 40 slides, each presenting one pair of words: Russian word and pseudoword. All Russian words were concrete nouns with frequency index from 8.5 to 22.8 per million, consisting of 7 letters. Pseudowords were made in Wuggi program [10] on the basis of English words (e.g. “consike”, “remwoud”, “stalore”—all the words were amendable to English phonetical rules). We checked “association strength” of the stimuli in preliminary research, which lead to exclusion of some words. Namely, if most of the subjects produced the same association to a word, it was excluded; as were excluded the words, where most of the subjects failed to provide associations. From the rest of the words we chose 40 pseudowords for the experiment.

Apparatus. The slides were presented on 19″ computer screen, in Courier New font (monospace), font size = 48. Eye movements were recorded with EyeLink 1000 eye tracker (SR Research), monocular, desktop mount, 500 Hz, camera to eye distance was 55–65 cm.

Processing. The data were processed in SR DataViewer 1.11.1 and SPSS’19 programs.

2.2 Procedure

The experiment was applied individually and took approximately 60 min. Before starting the experiment, calibrations were performed. The instructions were given both orally by the experimenter and written on the screen. Subjects were told that during the experiment they would need to learn as many words of a pseudo-foreign language as they could.

The slides were presented on a computer screen, and eye movements were recorded during the presentation. The experiment included 10 series (2 test series and 8 experimental series). Each series comprised 4 slides with word pairs (presented randomly). Presentation time was 5 s per slide. The slides were separated by blank screen with fixation cross in the middle (presented for 1 s). Half of the series had Russian-pseudoword presentation order (a Russian word was presented to the left of a pseudoword), in the other half presentation order was reverse—a pseudoword was presented to the left of a Russian word.

30 s after each series, subjects were given forms on which to provide pseudo-word equivalents to Russian words. Then subjects were given correct answers and they had to provide a free report of the memorizing process (Fig. 1).

Fig. 1
figure 1

The general scheme of the experiment

Full size image

2.3 Measures

Qualitative recall measures. We rated recall score on a scale from 1 to 14: letter written—1 point (max-7), letter in the right place—1 point (max-7).

Quantitative characteristics of processing and recall

Defining word memorization techniques. We elicited learning techniques on the basis of subjects’ post hoc report. In 12.5 % cases (124/992 trials) subjects did not provide the report. All other cases could be attributed to either graphical, or phonological, or semantic techniques, according to the following descriptions:

  • If the use of graphical technique was reported, Ss emphasized specific letters and their graphical features (e.g. “Each word had high-register letters…”). The method was used in 4.03 % cases.

  • If the use of the phonological technique was reported, it was described with such phrases as “I just read the words…” or “It rhymed with “pore”…”—the accent was on phonological features. The method was reported relatively often: 28.12 % trials.

  • The semantic technique was the one most commonly reported: 55.35 % trials. It was characterized either by visualization (“I imagined тeтpaдь (notebook), on which was written consike (pseudo-word)…”) or a chain of associations (for instance, auditory association to the pseudo-word (e.g. consike-кoнь (kon’, a horse) resulted in the report, “I imagined a horse eating a notebook…”). The latter has been described by Atkinson as the “keyword method” [1].

These techniques are described according to the “depth” of processing involved, as defined in the levels of processing theory [3, 27].

Analyzing recall mistakes. The mistakes were also classified according to the levels of processing:

  • “graphical” (mixing up letters similar in writing, e.g. “consile” instead of “consike”);

  • “phonological” (mixing up similarly read letters according to reading rules in the English language, e.g. “konsike” instead of “consike”);

  • “semantic” (mistakes in associations, e.g. if a keyword was “kon” (a horse), the mistake could be e.g. “ovessike” instead of “consike”Footnote 2);

  • “other” (mixed up words, missed letters, etc.).

Eye movement characteristics. In total 992 trials (878 valid) were recorded. Fixation count, dwell time on Russian and pseudoword AOIs, transition count (number of “switches” between the words), regressive eye movements within AOIs were analyzed.

3 Results and Discussion

The effect of conscious use of learning techniques was revealed: F(4, 992) = 12.81, p < 0.01 (recall score in the trials with no report of the technique was significantly lower than with the use of any learning method). The coefficient of contingency between the techniques used and the mistakes made was 0.316 (p < 0.01).

Significant distinctions were found in the recall score depending on the techniques used: F(4,878) = 17.1 (p < 0.01) (see Fig. 2). The recall score increased with the “depth” of processing involved: when graphical technique was reported, the mean recall score was 7.23/14 (std. dev. = 5.72); for the phonological—8.69/14 (std. dev. = 5.47); and for the semantic—10.22/14 (std. dev. = 4.50). The high recall score in methods including a visual component, as compared to “mechanical” methods, is consistent with the results of Paivio [18] and Atkinson [1].

Fig. 2
figure 2

Mean recall score (on the scale from 1 to 14) in the use of different techniques (mean values are given below the strategies, mean errors are marked as error bars)

Full size image

3.1 Eye Movement Characteristics in Different Learning Techniques

We excluded the trials where no technique was reported from data processing (leaving 770 trials). Significant differences in eye movement characteristics were found in mnemonic techniques (see Table 1).

Table 1 Eye movement characteristics in mnemonic techniques
Full size table

The graphical technique was characterized by a higher fixation count and regressive eye movements on the pseudo-words and a moderate amount of transitions between AOIs (see Fig. 3). Such an eye-movement pattern can be explained by “fractional” perception of the words with emphasis on specific letters.

Fig. 3
figure 3

S’s eye movements (graphical technique used): the circles represent fixations (duration of the fixations are given above the circles and also correspond to the radii), the arrows represent saccades (the order numbers of the saccades are given above the arrows)

Full size image

When subjects reported the use of the phonological method they made more transitions between AOIs, while making a minimal amount of fixations and regressive eye movements (see Fig. 4). This may be due to memorising phonetically.

Fig. 4
figure 4

S’s eye movements (phonological technique used): the circles represent fixations (duration of the fixations are given above the circles and also correspond to the radii), the arrows represent saccades (the order numbers of the saccades are given above the arrows)

Full size image

When using the semantic method of memorization, few transitions between Russian and pseudo-words and a relatively large number of fixations on pseudo-words were observed (see Fig. 5). This may be due to the fact that association-based memorizing is supposed to focus on the semantic level of information processing, which leads to a relatively rapid memorization of Russian words and, consequently, a longer dwell time on pseudo-words. This asymmetrical distribution of attention does not lead to recall mistakes associated with the formation of ties with the Russian word: if the subject has completely or partially reproduced the pseudo-word, in the majority of cases (438 of 519) it has been properly related to the Russian word.

Fig. 5
figure 5

S’s eye movements (semantic technique used): the circles represent fixations (duration of the fixations are given above the circles and also correspond to the radii), the arrows represent saccades (the order numbers of the saccades are given above the arrows)

Full size image

Thus, three eye-movement patterns, characteristic of the three techniques defined, were revealed. Our next step was to check whether similar eye-movement patterns can be distinguished irrespective of the techniques. We used cluster analysis to categorize the experimental series on the basis of S’s eye movement characteristics, namely, transition and dwell time on native and foreign words. The initial 3-cluster solution which we used did not yield any results that we could match with the techniques. Therefore we opted for a 2-cluster solution, which is given in the details below (see Table 2).

Table 2 Cluster analysis results
Full size table

The first cluster, as opposed to the second cluster, included a series with higher transition count, shorter fixation duration on foreign words, and longer fixation duration on native words. The coefficient of contingency between the clusters and the techniques that the subjects used was 0.12 (p < 0.01). The source of distinctions was in the phonological and the semantic method (See Table 3).

Table 3 The distribution of the trials between clusters according to the learning strategies used
Full size table

Therefore the distinguished eye movement patterns are partly verified by cluster analysis, however, further data has to be collected to make predictions of the learning method based on eye movements.

3.2 Eye Movement Characteristics Depending on the Localization of the First Fixation

We varied the order of presentation to provoke an even distribution of the first fixations between a Russian word and a pseudo-word. Eye movement characteristics in different learning strategies depended on the localization of the first fixation in a way that differed from the results already reported.

When graphical and phonological techniques were reported, the difference in transition count depending on the localization of the first fixation was not significant. This can be attributed to the lack of emphasis on semantic processing. Since the main focus was on memorizing the visual characteristics of the foreign words (in the graphical technique), and remembering the words by their phonetic link with the Russian word (in the phonological technique), it was not so critical which word was the first to be fixated.

When the semantic technique was used and the first fixation was made on the pseudo-word, the number of “transitions” was significantly higher—2.8 (1.1), than when the first fixation was made on the Russian word—2.3 (1.3); F = 27.9; p < 0.01. If we consider the process of memorizing foreign words in terms of inclusion of a new form into an existing logogen (in the terms of Morton [15]), this trend can be explained by the fact that when the learning started with the Russian word, the meaning was already given and it did not require any rehearsal, while in the opposite case the subject saw the new form first and then its meaning, and after that had to go back to the pseudo-word, as it was the word that had to be remembered.

4 Conclusions

We found and described three strategies of memorizing of new SL vocabulary units: the Graphical strategy, the Phonological strategy and the Semantic strategy. These strategies were manifested in mnemonic techniques, reproduction errors and distinctive eye movement patterns of the subjects. This is in accordance with the levels of processing theory [3, 27].

The results obtained show the influence of the conscious use of mnemonic techniques in recall activity—recall score increases with increasing “depth” of processing. Recall was significantly better when the technique included a visual component compared to “mechanical” methods of memorizing, which accords with the results of Paivio [18] and Atkinson [1].

The techniques used for non-contextual foreign vocabulary learning are reflected in eye movements. When the use of graphical technique was reported, a higher fixation count was observed, and more regressive eye movements on the pseudo-words and a moderate amount of transitions between AOIs were made. Such an eye-movement pattern can be explained by “fractural” perception of the words. When the use of the phonological technique was reported, subjects made more transitions between AOIs, while making a minimal amount of fixations and regressive eye movements. This may be due to learning the words by their phonetic link. When using the semantic method of memorization, few transitions between Russian and pseudo-words and a relatively large number of fixations on pseudo-words were observed.

The recall score was connected with the use of a conscious learning method and, in particular, the use of the semantic method, which scores higher on the “depth” of semantic processing. Eye movements in this technique were connected with the localization of the first fixation. The acquired data open up prospects for the evaluation of online learning strategies.