2.1 Theoretical background

DDL is fully compatible with usage-based second language acquisition (SLA) theories that conceive of language as an open-ended dynamic system that emerges by way of probabilistic bottom-up abstraction rather than a fixed system that follows categorical top-down rules (Ellis, Reference Ellis2014). One of the most important tenets of usage-based approaches is inseparability of grammar and lexis, conceptualized by Ellis (Reference Ellis2014: 399) as follows:

The second theoretical principle behind DDL is Schmidt’s (Reference Schmidt1990) noticing hypothesis, which posits that some level of learner awareness of the target L2 construction is necessary for it to be learned. In order to attract learners’ attention to L2 targets, specific instructional methods have been proposed, notably input enrichment and input enhancement. Input enrichment, or input flood, is increase of frequency of the target feature in the input (Trahey & White, Reference Trahey and White1993) and input enhancement refers to making the target feature more salient, for example with typographical means such as bolding, underlining, or color marking (Sharwood Smith, Reference Sharwood Smith1993). DDL is inherently conducive to both teaching techniques as corpora can supply a large number of attested language samples containing the target construction (input enrichment) and these samples can be retrieved from corpora with the help of concordance tools in form of stacked concordance lines with the target construction highlighted and centered (input enhancement). Although input enrichment and enhancement had been originally proposed as rather implicit instruction methods (i.e., those facilitating learner awareness at the level of subconscious noticing), empirical SLA research has shown that implicit instruction is often insufficient (especially for non-salient L2 targets) and that more explicit methods (i.e., those facilitating learner awareness at the level of conscious understanding) are necessary (Ellis, Reference Ellis2005; Sharwood Smith, Reference Sharwood Smith2013; Spada & Tomita, Reference Spada and Tomita2010). The specific method that has been adopted in most DDL interventions can be characterized as explicit inductive within the implicit–explicit/inductive–deductive taxonomy (DeKeyser, Reference DeKeyser2003: 314). In this method, also termed “discovery learning” (Bernardini, Reference Bernardini2002), rich and enhanced input (e.g., concordance lines) serves as material for learners’ noticing and analysis of language patterns. As a result, learners arrive at generalizations about the usage of the target constructions, which, in turn, lead to the complexification and expansion of their L2 knowledge (Flowerdew, Reference Flowerdew2015).

The third principle is that of learner autonomy, the development of which has been repeatedly pointed out as an important long-term benefit of DDL (e.g., Chambers & O’Sullivan, Reference Chambers and O’Sullivan2004). However, it has also been acknowledged that this development should be mediated through teacher and peer assistance. The adoption of the principle of the mediated nature of language development from sociocultural theory (see Flowerdew, Reference Flowerdew2015) has led to the modification of the inductive discovery method to the so-called guided induction approach. This method, originally proposed by Herron and Tomasello (Reference Herron and Tomasello1992) in the area of general language teaching, has recently been adopted by many DDL teachers-researchers (Flowerdew, Reference Flowerdew2009, Reference Flowerdew2015; Frankenberg-Garcia, Reference Frankenberg-Garcia2014; Huang, Reference Huang2008; Yoon & Jo, Reference Yoon and Jo2014). Smart (Reference Smart2014: 187) describes guided induction as follows:

It must be noted that guidance implies not only teacher but also peer scaffolding (cf. “interactive task”), when learners collaboratively work on corpus tasks and assist each other in the inductive discovery process (Flowerdew, Reference Flowerdew2015; Kennedy & Miceli, Reference Kennedy and Miceli2001).

In sum, the abovementioned theoretical underpinnings inform the main pedagogical principles behind DDL, some of which stand in stark contrast to widely spread conventional teaching techniques as summarized by Flowerdew (Reference Flowerdew2015: 15–16): (1) a lexico-grammatical approach as opposed to the separation of lexis and grammar into different curricular modules; (2) reliance on naturally occurring attested language as opposed to concocted textbook examples; and (3) guided inductive methods based on rich and enhanced input as opposed to rule-based deductive methods.

2.2 Empirical DDL research

Although DDL research is still a young area of inquiry, it has accumulated enough empirical studies for several research syntheses (Boulton & Pérez-Paredes, Reference Boulton and Pérez-Paredes2014; Chambers, Reference Chambers2007; Leńko-Szymańska & Boulton, Reference Leńko-Szymańska and Boulton2015) and the first meta-analysis (Cobb & Boulton, Reference Cobb and Boulton2015). Having integrated the results of quantitative studies published up to 2012, Cobb and Boulton showed large effect sizes for both learning gains with DDL methods and their superiority to conventional teaching methods. The overwhelming majority of L2 foci in these studies included English collocations (frequently co-occurring patterns of words) taught with DDL methods to learners in different countries and instructional settings. Most of the earlier studies explored the so-called “hard” (Gabrielatos, Reference Gabrielatos2005) version of DDL, where learners searched corpora online on their own with minimal guidance from their teachers. However, this version was soon recognized as not equally feasible for all instructional settings. Online corpus searches by learners turned out to be challenging for many students and teachers alike due to high cognitive task demands, lack of technological expertise and support, or simply absence of computer-equipped classrooms (Farr, Reference Farr2008; Tian, Reference Tian2005). This hurdle has led to the emergence of numerous DDL modifications that can be positioned on a “cline” from hard to soft versions (Mukherjee, Reference Mukherjee2006), depending on the medium (computer or paper) and task difficulty (open-ended and controlled, addressing variable or fixed rules). The number of studies devoted to “softer” DDL versions remains small but it has been recently growing, as evidenced by the overview below.

The effectiveness of paper-based DDL interventions, in which learners work with concordances selected and printed on worksheets by their teachers, has been investigated in a number of exploratory and experimental studies (see Boulton, Reference Boulton2010, for an overview). All studies that employed quantitative methods showed that paper-based DDL led to significant gains in learners’ knowledge of a variety of L2 English targets (e.g., phrasal verbs, connectors, and passive voice constructions), and that DDL was either as effective as or better than conventional rule-based methods. Furthermore, quantitative studies that compared computer-based and paper-based DDL found no difference in the effectiveness of these two methods (Boulton, Reference Boulton2012; Vyatkina, Reference Vyatkinain press). Boulton (Reference Boulton2008, Reference Boulton2009, Reference Boulton2010, Reference Boulton2012), in particular, has shown that “taking the computer out of the equation” (Boulton, Reference Boulton2010: 234) in paper-based DDL does not compromise its benefits for language instruction. Several studies have also shown that paper-based DDL works at both higher and lower proficiency levels (Koosha & Jafarpour, Reference Koosha and Jafarpour2006; Tian, Reference Tian2005; Yoon & Jo, Reference Yoon and Jo2014). Boulton (Reference Boulton2008, Reference Boulton2009, Reference Boulton2010, Reference Boulton2012) has conducted systematic paper-based DDL research with first language (L1) French university learners of English at relatively low levels of proficiency (levels A2–B1 of the Common European Framework of Reference, or CEFR; Council of Europe, 2001) despite up to eight years of instruction. Boulton showed that DDL helped such learners to acquire certain linguistic targets that were often impermeable to traditional instruction methods. A number of studies focused on verb-preposition collocations, which are the target of the present study. These constructions are notoriously difficult for learners with any L1-L2 background because of form-meaning mapping mismatches between languages (Kennedy & Miceli, Reference Kennedy and Miceli2001; Nesselhauf, Reference Nesselhauf2004). DDL studies that have targeted English verb-preposition collocations either alongside other foci (Boulton, Reference Boulton2010; Frankenberg-Garcia, Reference Frankenberg-Garcia2014) or as a sole instructional focus (Koosha & Jafarpour, Reference Koosha and Jafarpour2006) have shown positive DDL effects. It must be noted that most DDL interventions have been administered by the researchers themselves with two notable exceptions: Yoon (Reference Yoon2008), who closely worked with an instructor who administered computer-based DDL to his students over a longer period; and Boulton (Reference Boulton2010), who reports on a successful short paper-based DDL intervention administered by several teachers.

Several recent studies have fleshed out specific features of DDL conducive to language learning. Frankenberg-Garcia (Reference Frankenberg-Garcia2014) demonstrated that learners who worked with multiple concordance lines for English collocations achieved greater gains in L2 production than learners who worked with one concordance line, worked with definitions of the target structures, or were in a control group. Smart (Reference Smart2014) showed that for English passive constructions, the guided inductive DDL group outperformed two deductive groups, one which worked with corpus-informed materials and another with constructed examples from textbooks. The only study (conducted in the computer-based DDL context) that explicitly singled out the specific knowledge effect is Chan and Liou (Reference Chan and Liou2005). They found that learners with lower entry-level collocational knowledge made greater gains than learners with higher entry-level knowledge. However, the students in Chan and Liou’s study had had some previous knowledge of the focal collocations, which is in line with virtually all DDL research that has focused on “‘known’ but error-prone items” (Boulton & Pérez-Paredes, Reference Boulton and Pérez-Paredes2014: 124) that have been claimed to be more amenable to DDL interventions than completely new items (Cobb, Reference Cobb1999; Nesselhauf, Reference Nesselhauf2004). On the other hand, informal feedback from Boulton’s (Reference Boulton2008) participants suggested that they had had no previous knowledge of the target items (two phrasal verbs), which they successfully learned with paper-based DDL. Nevertheless, empirical comparisons of how DDL fares in teaching previously learned versus new items (a task especially relevant at early stages of instruction) are yet to be conducted.

The results of the research reviewed above thus suggest that the primary DDL benefits are richness of input and the use of a guided inductive approach, that both computer-based and paper-based DDL can be equally effective provided that they follow these principles, and that paper-based DDL works even for lower proficiency learners. However, a number of research gaps still remain, some of which will be addressed in the present study.

2.3 Remaining gaps and the goals of this study

Despite the positive findings discussed above, Mauranen’s (Reference Mauranen2004: 208) call, issued a decade ago, is still relevant and urgent: “to make a serious contribution to language teaching, corpora must be adopted by ordinary teachers and learners in ordinary classrooms”. One way to bring this project to fruition is to promote paper-based DDL that is more feasible for “ordinary” settings. However, in addition to the general scarcity of studies that compare the effectiveness of paper-based DDL and non-DDL methods (see review above), there are a number of design limitations that have yet to be addressed. The first research gap concerns the target language focus. So far studies have been almost exclusively limited to English. This, in turn, has led to primary attention directed at lexical linguistic targets and analytical grammatical constructions (e.g., passive and subjunctive verb forms), whereas morphologically complex targets (e.g., nominal inflection), which are largely irrelevant to English, have flown under the radar. A case in point is verb-preposition collocations, which in English essentially belong to the lexical domain as they consist of a content word and a functional word. In contrast, German verb-preposition collocations represent a complex lexico-grammatical construction because either the verb or the preposition assigns grammatical case to the subcategorized noun phrase indexed by inflectional markers on that phrase. The German nominal inflectional paradigm represents a salient case of morphological complexity (Pallotti, Reference Pallotti2015), and verb-preposition-case collocations have been shown to be difficult for L2 German learners at different proficiency levels (Baten, Reference Baten2011; Vinagre & Muñoz, Reference Vinagre and Muñoz2011). At the same time, explicit teaching has been shown to be necessary to attract learners’ attention to inflectional morphology – an abstract, low-salience grammatical feature (DeKeyser, Reference DeKeyser2003; Ellis, Reference Ellis2005). Therefore, it is worth exploring whether DDL as an explicit inductive method may fare better than traditional explicit deductive methods to achieve this goal.

Furthermore, the empirical DDL research field is in urgent need of replication studies, as any classroom research is by nature small scale and quasi-experimental with intact classes functioning as participant groups. Partial replication studies can enhance the generalizability of classroom research findings (Chun, Reference Chun2012; Porte, Reference Porte2012), but such studies are still rare in DDL (see, however, Frankenberg-Garcia, Reference Frankenberg-Garcia2014). This study aims to take this avenue by partially replicating two recent DDL studies. The first study is Boulton (Reference Boulton2010), who compared the effectiveness of brief interventions for teaching a number of English collocations to L1 French EFL (English as a foreign language) students at the A2-B1 proficiency levels. The study showed that the paper-based DDL method was marginally better than the non-DDL dictionary-based method, with more students improving their scores on more items, and that both methods were better than no instruction. The study thus argued in favor of applying paper-based DDL along with more traditional methods in teaching some lexico-grammatical targets to low-proficiency students. The second study is Vyatkina (Reference Vyatkinain press), who compared the effects of computer-based and paper-based DDL for teaching German verb-preposition collocations to L1 English learners at the B1 proficiency level. That study showed that both methods were equally effective, with higher and longer lasting gains for controlled production tasks (gap filling) than for free production tasks (sentence writing). The present study is similar to Boulton (Reference Boulton2010) in that it compares the effects of a paper-based DDL and a non-DDL method administered by regular teachers as brief, narrowly focused interventions embedded within a non-DDL curriculum and without prior DDL training of the participants. It also replicates Vyatkina (Reference Vyatkinain press) in that it focuses on learners of German and targets German verb-preposition collocations as complex lexico-grammatical constructions. The novel features of the present study are that its participants have very low L2 proficiency levels (CEFR A2 and below), that it assigns intact classes to different treatment conditions (unlike Boulton and Vyatkina who administered different treatments to the same cohort of learners), that it separates lexical and morpho-syntactic learning, and that it compares the DDL effects for teaching completely new versus previously learned collocations. The broader purpose of this study is, similar to Boulton’s (Reference Boulton2010: 541), to “counter a number of frequent objections to DDL and contribute to greater awareness of its potential”.

3.1 Research questions

The study explores the following research questions:

1. 1. Does learner lexical and morpho-syntactic knowledge improve following focused instruction (as demonstrated by written performance on a closed gap-filling task)?

2. 2. Are the gains higher following the DDL or the non-DDL treatment?

3. 3. Are there interactions between treatment, course level, and linguistic items (lexical or morpho-syntactic)?

3.2 Participants and institutional setting

The study was administered at a large public North American university. Participants were recruited from the third- and fourth-semester German classes in a four-semester-long program that fulfills the foreign language requirement for certain majors. The classes met three times per week for 50 minutes. This is a multi-section program in which all sections of the same course follow a uniform syllabus and use the same textbook. The instructional approach combines a communicative approach with focus-on-form activities, and all courses devote an approximately equal amount of time to speaking, writing, reading, listening, vocabulary, grammar, and cultural learning. All syllabi also have a substantial learning-with-technology component including an electronic workbook and biweekly computer lab meetings, mostly devoted to searching German websites for cultural information, but without a DDL component. All seven experimental classes in this study were taught by graduate student instructors under the researcher’s supervision. Some instructors had basic knowledge of corpora and DDL from their graduate coursework (see Vyatkina, Reference Vyatkina2013) but none of them used DDL in teaching the focal classes.

Altogether, 88 students participated in the two iterations of the experiment. All of them had American English as their L1. The average age was 21 (18–35 range). Gender was not considered a variable in this study, however it can be mentioned that although the proportion of females and males in each class differed, the overall distribution was balanced (43 females and 45 males). The L2 German proficiency of the participants was fairly low. Students who enroll in the first semester of this program have no or almost no knowledge of German. Some of them progress through all four semesters in the program while others join it at later time points via a placement test. To obtain a more general measure of the participants’ L2 proficiency, a standardized German proficiency test was administered at the end of the experimental semester. All fourth-semester classes took an official diagnostic test which was administered by the onDaF Institute in Bochum, Germany (www.ondaf.de) and proctored at the researcher’s institution. Participants take this online cloze test over 40 minutes and are then automatically placed within CEFR bands (Eckes & Grotjahn, Reference Eckes and Grotjahn2006). The results showed that approximately two thirds of all fourth-semester students reached the A2 level by the end of the semester, one third did not reach it, and only two participants reached the B1 level. Since the onDaF test only measures proficiency at or above the A2 level, it was not considered meaningful to test third-semester students. Overall, although proficiency is not considered a variable in this study, it is safe to state that both third- and fourth-semester students in the focal instructional program were generally at or below the A2 CEFR level, i.e. had roughly low-intermediate L2 proficiency.

Importantly, third- and fourth-semester participants differed not only in their course level but also in their entry-level knowledge of the target items. The third-semester students were explicitly taught them for the first time, whereas the fourth-semester students were reviewing material taught in the previous semester. Although previous incidental exposure to the target items by the third-semester students cannot be excluded, the bulk of the engagement of this student population with their L2 is limited to the classroom. As far as the fourth-semester groups are concerned, the overwhelming majority of the students progressed to their fourth-semester course immediately after the third-semester course, which means that they were first exposed to the target items at the same time and in the same fashion during the previous semester. As the intervention was administered during a spring semester, only a relatively short (one-month-long) winter break separated the semesters, which does not allow much time for knowledge attrition or additional out-of-class exposure to the L2 (e.g., traveling to German-speaking countries, which none of the participants undertook). Therefore, most of the residual knowledge of the target items in the fourth-semester groups can be confidently attributed to the instructional effects from the previous semester. The only exception is three participants who enrolled in the course via a placement test, and who therefore did not follow the same instructional sequence as other participants. However, they showed an entry-level knowledge of the target items similar to other participants and thus were included in the study.

3.3 Target items

The intervention focused on eleven verb-preposition collocations (Appendix 1) from the regular textbook used in the first three semesters in this program (Di Donato, Clyde & Vansant, Reference Di Donato, Clyde and Vansant2012: 359). In ten of these collocations, prepositions are not congruent in German and English: For example, to wait for is equivalent to warten auf, although the prototypical translation of the preposition for is für. Furthermore, the German noun phrase or the pronoun following the preposition carries an obligatory gender, case, and number marker (e.g., Ich warte auf meinen Bruder/ihn [masculine, accusative, singular] – I am waiting for my brother/him). However, if the pronoun refers to an inanimate object, German uses pronominal adverb-preposition contractions, the so-called da-compounds, instead (e.g., Ich warte darauf – I am waiting for that), in which no inflectional markers are present. Furthermore, German pronominal adverbs (e.g., davon, dabei) are extremely frequent, not genre-restricted, and can combine with most prepositions, unlike their very infrequent and genre- and item-restricted English counterparts (e.g., thereof, thereby). German da-compounds are considered an important part of active vocabulary for learners and are typically taught in conjunction with prepositional verbs along with prepositional phrases. Needless to say that, being an important instructional focus due to the sheer frequency of verb-preposition collocations in language usage, such lexically, grammatically, and semantically complex constructions present considerable difficulties to learners, although typical syllabi and textbooks allocate only one to two lessons to introducing them.

3.4 Procedures

There were two iterations of the experiment. In the first iteration, two sections of the third-semester course and three sections of the fourth-semester course participated. The experimental sections were assigned to four conditions based on treatment (D=DDL; T=textbook) and course level (L=low, i.e. third semester; M=mid, i.e. fourth semester). Only the data of the students who agreed to participate and took both the pretest and the posttest were included in the study. The DL (n=15) and TL (n=13) sections were taught by the same instructor. Two fourth-semester sections taught by a different instructor were assigned the TM condition due to the low number of students (combined n=16). Finally, the third fourth-semester section, taught by yet another instructor, was assigned the DM condition (n=13). It must be noted that the DM section followed a syllabus different from the TM sections as it constituted a different course (introduction to German for the professions). However, since the intervention was administered on the tenth day of the semester in fourth-semester classes, no noticeable difference in instruction had yet taken place. Also, since the regular instructor of the DM section was out of town on the day of the experiment, the researcher administered the intervention. Since she was known to the students as program coordinator, and the regular instructor was still new to the students, the intervention did not cause any considerable disruption to the regular instructional flow. After the first iteration of the experiment, the researcher decided to replicate the study with another cohort of third-semester students with slight design changes (see explanations under Results below). Two sections taught by an instructor who did not participate in the first iteration of the study were assigned to the DL2 (n=16) and TL2 (n=15) conditions.

The intervention in the third-semester classes was administered at the time designated for the target items in the course syllabus (after about one third of the semester was over), and in the fourth-semester classes as part of the start-of-the-semester review. Prior to the intervention, the researcher met with all participating instructors individually for about 30 minutes, discussed the interventions, gave them packets with detailed lesson plans, test sheets, and worksheets, and answered their questions. In the T-groups, the procedure followed the regular syllabus (barring the pretest and the posttest), so the instructors did not have to change anything in their teaching. Regarding the D-groups, instructors showed much interest in trying out the new teaching method. The ten-minute-long pretest was administered at the end of the class immediately preceding the intervention class. The homework assignment in all groups was unrelated to the target items. The intervention, administered during the next class, lasted for 40 minutes and the class concluded with a ten-minute-long posttest. Both tests were announced to the students as ungraded quizzes. Additionally, all participants filled out a brief electronic personal and language learning background questionnaire during one of their regular computer lab meetings.

The test instruments were paper worksheets with fourteen sentences from the DWDS corpus: a large, freely and publicly available corpus of contemporary German (www.dwds.de). Ten sentences contained the target verbs and four sentences contained control items (see Boulton, Reference Boulton2010) – verb-preposition collocations not taught in the intervention. Only ten target constructions were tested because two out of the eleven collocations contained the same verb with different prepositions that indicated different meanings. To remove this additional difficulty, only one collocation of these two was tested. The verb was followed by a preposition and a noun phrase with a definite article in half of the sentences and by a da-compound in the other half. Only singular nouns were used and the gender of each noun was indicated in parentheses so that participants had to only think about case while working on the grammatical items. The target prepositions, articles, and da-compounds were blanked out and learners had to fill the gaps. A model with a non-target prepositional verb introduced each part of the test. The verbs that were used with prepositional phrases in the pretest were used with da-compounds in the posttest and vice versa. During scoring, one point was given for each correctly supplied lexical item (preposition or da-compound), and one point for each correctly supplied grammatical item (article) after correct prepositions.

The difference between the 40-minute-long treatments was the following. The TL and TM groups followed the sequence of activities in the textbook (Di Donato et al., Reference Di Donato, Clyde and Vansant2012: 359–362). This lesson followed the typical deductive ‘triple P’ (Presentation-Practice-Production) model (e.g., DeKeyser, Reference DeKeyser1998). The target verb-preposition-case collocations were first presented as a list with English translations. A practice drill followed, and then da-compounds were introduced in contrast with preposition-pronoun collocations with a few examples. Next, students practiced using da-compounds in individual, pair-work, and whole-class drills. The sequence concluded with an oral pair-work exercise in which students exchanged questions and answers about everyday topics using the target constructions. In contrast, the DL and DM groups worked with DDL worksheets containing five to seven concordance lines for each collocation (Appendix 2; see also Vyatkina, Reference Vyatkina2015) copied and pasted from the DWDS corpus. The intervention followed the inductive ‘triple I’ (Illustration-Interaction-Induction) model proposed by Carter and McCarthy (Reference Carter and McCarthy1995) and extended by Flowerdew (Reference Flowerdew2009) to the guided induction model with an additional ‘Intervention’ step between Interaction and Induction. The instructor in each D-class briefly introduced the idea of a corpus as a rich repository of language usage examples and distributed the worksheets. Then, she instructed the students to find patterns in the concordance lines and modeled that with one focal verb. The students then worked in pairs discussing the verb-preposition-case patterns and wrote individual examples for each verb. Next, the instructor discussed the results with the whole class, making sure that everybody arrived at the right pattern. A similar procedure was followed with concordance lines for verb-preposition-pronoun collocations and da-compounds. The lesson concluded with a Q&A oral pair-work exchange similar to that in the T-groups. Therefore, the lessons in all groups were designed as in Boulton’s (Reference Boulton2010: 547) study: “The intention was thus that all students should come away with essentially the same final information; the main differences lay in the way it had been reached and the materials from which it was derived.”