Transfer of collostructions: the case of causative constructions

1 Introduction

Since its development twenty years ago by Stefan Gries and Anatol Stefanowitsch (cf. Gries and Stefanowitsch 2004a, 2004b; Stefanowitsch and Gries 2003), collostructional analysis has been applied to several constructions, languages and fields of linguistics. It has made it possible to measure the degree of association between constructions and words within these constructions, thus pointing to the interaction that can exist between them, but also contributing to a better description of the meaning of constructions.

This article applies collostructional analysis to the investigation of periphrastic causative constructions in English and French, and brings together the fields of contrastive linguistics and second language acquisition research, through the use of the Integrated Contrastive Model (Gilquin 2000/2001; Granger 1996). It seeks to uncover possible transfer effects in the use of English periphrastic causative constructions by French-speaking learners of English, focusing on the collostructional preferences in the non-finite verb slot. At the same time, the article is an illustration of how transfer can be investigated by combining what can be referred to as contrastive collostructional analysis and contrastive interlanguage collostructional analysis.

2 Periphrastic causative constructions and their collexemes

Periphrastic causative constructions use analytical means to express that an entity X influences an entity Y in such a way that Y performs some act. They are typically made up of a causer (X), a causee (Y), a causative verb (e.g. cause, get or make in English) and an effect that takes the form of a non-finite verb complement. Examples for English and French are given in (1) and (2), which include the causative verbs make and faire . These two verbs will be the focus of this study because they are the main verbs used in periphrastic causative constructions and they are usually considered equivalent: faire is the most obvious translational equivalent of make , both in causative and other uses (although there are some differences between [X make Y V_inf] and [X faire V_inf Y], see Gilquin 2008).

While periphrastic causative constructions are characterized by many semantic, syntactic, lexical and stylistic features, one aspect that has been relatively under-researched and yet is crucial in describing these constructions is the phraseology of the non-finite verb. In Gilquin (2010: 137–139), a multinomial logistic regression was used to predict ten periphrastic causative constructions in English from forty features extracted from corpus data (e.g. genre, animacy of the causer, tense of the causative verb, transitivity of the effect). The lemma of the effect turned out to be the most discriminative feature when applying both backward and forward selection.

Collostructional analysis makes it possible to highlight, and also quantify, associations between causative constructions and verbs in the effect slot. A multiple distinctive collexeme analysis of the ten English causative constructions referred to above thus pointed to verbs like feel, laugh, look and think as being very strongly attracted to the [X make Y V_inf] construction as compared to the other nine constructions (Gilquin 2006) and showed that these ‘collexemes’ belonged to the more general category of non-volitional verbs.

Although such associations are expected to be part of native speakers’ constructicons (i.e. their mental repertoire of constructions), this is not necessarily the case for language learners, who may have difficulty approximating the native collostructional preferences of causative constructions (see Gilquin 2012). There are at least three possible reasons for this. First, learners’ lack of exposure to the target language may have prevented them from storing these preferred associations in their constructicons. Second, preferences in the non-finite verb slot of causative constructions are hardly ever described in pedagogical tools or even in the specialized literature (see Gilquin 2006, 2010), so that learners are unlikely to have heard about these. The third reason, which is the focus of this article, is that the collostructional preferences of causative constructions may differ in learners’ mother tongue and in the target language, thus potentially leading to cases of transfer from the mother tongue. Here, I investigate the possible role of transfer in French-speaking learners’ collexemes in the non-finite verb slot of the [X make Y V_inf] causative construction. To do so, I use the Integrated Contrastive Model, which is outlined in the next section.

3 Transfer and the Integrated Contrastive Model

In second language acquisition research, transfer describes learners’ tendency to reproduce in the target language forms or patterns that come from another acquired language, including the mother tongue (see, e.g., Skehan 2008). When forms/patterns in the two languages differ from each other, transfer leads to the use of inappropriate forms/patterns in the target language and is then called negative transfer. Transfer can concern any aspect of language, including lexical and phraseological aspects. Using experimental data, Matthys (2015) demonstrates the existence of ‘collostructional transfer’ in the dative alternation among Dutch learners of English, meaning that the learners’ collostructional preferences in English are partly influenced by the collostructional preferences of the equivalent constructions in Dutch.

Transfer is at the core of the Integrated Contrastive Model (ICM; Gilquin 2000/2001; Granger 1996), a corpus-based model that involves the combination of contrastive analysis and contrastive interlanguage analysis to identify possible cases of transfer (see Figure 1). Applied to English causative constructions used by French-speaking learners of English, the model requires a contrastive analysis (CA) comparing causative constructions in the learners’ mother tongue (French) and in the target language (English). The comparison can be drawn on the basis of translation corpora (texts in the source language [SL] and their translations into the other language [TL]) and/or comparable corpora (original texts [OL] in the two languages). The aim is to reveal any similarities and differences between the French and English causative constructions. The contrastive interlanguage analysis (CIA) compares the English causative constructions produced by French-speaking learners (IL, i.e. interlanguage) with those produced by English native speakers (NL, i.e. native language) and/or by learners from other mother tongue backgrounds (e.g. German- and Chinese-speaking learners; IL). The aim is to bring to light any uses of English causative constructions that are distinctive of French-speaking learners.

Figure 1:

The Integrated Contrastive Model (taken from Gilquin 2000/2001: 100; based on Granger 1996: 47).

The assumption behind the ICM is that differences revealed by the CA could result in negative transfer and lead to non-native features that are distinctive of the learner population under study. Thus, French-speaking learners who transfer features of causative constructions from French to English in cases where French and English causative constructions behave differently will produce causative constructions exhibiting non-native features unlikely to be found among learner populations with different mother tongue backgrounds.

Depending on the starting point (CA or CIA), the ICM can be used in a predictive way, to predict learners’ difficulties on the basis of the CA, or in a diagnostic way, to explain the origin of problems emerging from the CIA. As indicated by the diverging arrows in Figure 1, features of learner language are not all explained by differences between the mother tongue and the target language, in the same way as differences between the mother tongue and the target language do not necessarily lead to cases of transfer (they can help make predictions but these may not be borne out by the CIA).

In the next sections, the ICM is used to look for possible cases of collostructional transfer, by comparing the collostructional preferences of English and French causative constructions (contrastive collostructional analysis; Section 4), those of English causative constructions produced by English native speakers and by learners from French and other mother tongue backgrounds (contrastive interlanguage collostructional analysis; Section 5), and then bringing these two types of analyses together (Section 6).

4 Contrastive collostructional analysis of MAKE and FAIRE causative constructions

4.2 Main results

Below are the main results of the two independent simple collexeme analyses (Section 4.2.1) and those of the translation-based distinctive collexeme analysis (Section 4.2.2). As is usually the case with collostructional analysis, the collexemes are presented around classes of semantically related verbs. The focus is on classes that are distinctive of one language or the other.

4.2.1 Simple collexeme analyses

Figures 2 and 3, like all similar figures in this article, are plots produced by Coll.analysis 4.0 (Gries 2022). They show the verbs found in the [V_inf] slot of [X make Y V_inf] and [X faire V_inf Y], respectively (simple collexeme analyses).^[2] The verbs are organized along the y-axis according to their association with the construction (log odds ratio, henceforth LOR) and along the x-axis according to their frequency in the construction (logged co-occurrence frequency). In the left part of the plot, above the horizontal dashed line, are verbs that are associated with the construction but are relatively rare (e.g. reassume and shoulder in Figure 2, with just one occurrence each). Those in the upper left-hand corner are particularly strongly associated with the construction (e.g. boggle, which occurs only in a causative construction in the data). In the right part of the plot, above the horizontal dashed line, are verbs that are both associated with the construction and quite frequent (e.g. seem, with 25 occurrences). The verbs below the horizontal dashed line are not associated with the construction, having a negative LOR. The full lists of results can be found in Appendices 1 and 2, respectively.

Figure 2:

Log odds ratio and logged co-occurrence frequency of the collexemes of [X make Y V_inf] in an academic writing sample of the BNC.

Figure 3:

Log odds ratio and logged co-occurrence frequency of the collexemes of [X faire V_inf Y] in the academic writing component of Scientext.

Comparing the collexemes of the English and French constructions reveals one common verb class, namely verbs of (mental and sensory) perception. Collexemes representing this verb class include realize, think, notice, understand and reflect in English; prendre conscience (‘become aware’), remarquer (‘notice’), croire (‘believe’), comprendre (‘understand’) and penser (‘think’) in French. Some examples are given in (3) and (4).

(3)

Whether we can at the same time make people think explicitly about women is a more difficult question. (BNC-CMH-1225)

(4)

L’une des implications de cette étude est de faire prendre conscience aux enseignants de la subjectivité relative de leur comportement. (‘One of the implications of this study is to make teachers become aware of the relative subjectivity of their behaviour.’) (Scientext-454)

In English, [X make Y V_inf] also attracts copular verbs such as seem, feel, look and sound, as illustrated in (5).

(5)

There are many aids to make plots look more professional. (BNC-B16-795)

In French, several verbs of appearance are associated with [X faire V_inf Y], including ressortir (‘stand out’), émerger (‘emerge’), surgir (‘surge’) and naître (lit. ‘be born’), as exemplified in (6).

(6)

Le croisement de ces deux tableaux fait ressortir des disparités en apparence plus grandes. (‘Cross-referencing these two tables reveals [lit. makes stand out] seemingly greater disparities.’) (Scientext-617)

Interestingly, the verb appear and its French counterpart apparaître, which are both frequent and quite strongly associated with the causative construction, illustrate the difference between the class of copular verbs attracted to [X make Y V_inf] and the class of verbs of appearance attracted to [X faire V_inf Y]. While these verbs are obvious translations of each other, appear is mostly used as a copular verb, e.g. (7), whereas apparaître is mostly used to describe some entity becoming visible, e.g. (8).

(7)

For the time being let us remove two other misunderstandings which make the above conclusion appear paradoxical. (BNC-ANH-649)

(8)

Nos recherches ont fait apparaître des différences dans les pratiques des enseignants avec les élèves dits en difficulté. (‘Our research has revealed [lit. made appear] differences in teachers’ practices with so-called special needs students.’) (Scientext-594)

Another verb class attracted to [X faire V_inf Y] is that of verbs describing a change of state or location, with collexemes such as pivoter ‘pivot’, chuter ‘fall’, basculer ‘topple’, progresser ‘progress’, évoluer ‘evolve’ and varier ‘vary’, e.g. (9) and (10).

(9)

La référence à la pratique magistrale de l’explication fait basculer la réflexion. (‘The reference to the magisterial practice of explanation shifts [lit. makes topple] the thinking.’) (Scientext-617)

(10)

Les enfants ont été enregistrés dans deux situations conçues pour faire varier le degré de formalité. (‘Children were recorded in two situations designed to make the degree of formality vary.’) (Scientext-586)

4.2.2 Distinctive collexeme analysis

Figure 4 shows the results of the direct comparison between the [V_inf] slot of [X  make  Y V_inf] and [X faire V_inf Y] (translation-based distinctive collexeme analysis; see Appendix 3 for the full list of results). The verbs below the horizontal dashed line are more distinctive of [X make Y V_inf], whereas those above the line are more distinctive of [X faire V_inf Y]. Their degree of distinctiveness varies according to their position along the y-axis: the closer to the dashed line they are, the less distinctive; the more distant they are from the line, the more distinctive. The position of these collexemes along the x-axis indicates their (logged co-occurrence) frequency in the construction, from the least frequent on the left to the most frequent on the right.

Figure 4:

Log odds ratio and logged co-occurrence frequency of the distinctive collexemes of [X make Y V_inf] (bottom part) and [X faire V_inf Y] (top part), based on an academic writing sample of the BNC and the academic writing component of Scientext.

The results of the distinctive collexeme analysis partly confirm those of the two independent simple collexeme analyses. Some of the verbs with a relatively low LOR belong to the class of (mental and sensory) perception verbs, shown above to be common to the two constructions, e.g. understand, notice, realize, think (and their French equivalents). The LOR of appear is rather low too (0.85), which is due to the collexeme appear/apparaître being shared by the two constructions and to the lack of semantic distinction being made in the distinctive collexeme analysis between the copular use (typical of English) and the appearance use (typical of French).

Copular verbs are more distinctive of [X make Y V_inf] than of [X faire V_inf Y], e.g. feel, seem, look, sound, but the difference is especially striking for look, which is actually the most strongly associated collexeme in English (LOR = 27.21). In the case of feel, the contrast is less marked (LOR = 2.47), but this hides differences in usage between English and French. While in English feel is mostly used as a copula, cf. (11), the French equivalent sentir tends to occur as a verb of perception used transitively, cf. (12).

(11)

The family combines to make the daughter feel guilty and re-assume her task. (BNC-CE1-714)

(12)

Le Chêne montre déjà du sentiment et de la compassion, mais de cette compassion orgueilleuse par laquelle on fait sentir au malheureux les avantages qu’on a sur lui. (‘The Oak already shows feeling and compassion, but that kind of proud compassion by which one makes the unfortunate feel the advantages one has over him.’) (Scientext-617)

Several verbs of appearance and change of state or location turn out to be more distinctive of [X faire V_inf Y], such as the French equivalents of vary, emerge, evolve and progress (LOR of 24 or 25). The LOR for stand out is lower (2.08), but the collexeme is still distinctive of the French construction.

5 Contrastive interlanguage collostructional analysis of MAKE causative constructions

5.2 Main results

Below are the main results of the distinctive collexeme analysis comparing FrLE and NE (Section 5.2.1) and that comparing FrLE and GenLE (Section 5.2.2). The collexemes are presented around classes of semantically related verbs, with a focus on classes that are distinctive of one variety or the other.

5.2.1 Comparison of French learner English and native English

Figure 5 shows the results of the distinctive collexeme analysis of the [V_inf] slot of [X  make Y V_inf] in FrLE and NE (see Appendix 4 for the full list of results). The verbs above the line are more distinctive of FrLE, whereas those below the horizontal dashed line are more distinctive of NE. Their position along the x-axis shows their (logged co-occurrence) frequency in the construction. The plot reveals a number of collexemes that are close to the horizontal dashed line (e.g. feel, think, become). These are collexemes that are only very slightly distinctive of one variety or the other, having LORs lower than 2. The collexemes at the top and at the bottom of the plot are verbs that occur in one variety only and are therefore highly distinctive of that variety, having LORs higher than 20.

Figure 5:

Log odds ratio and logged co-occurrence frequency of the distinctive collexemes of [X make Y V_inf] in FrLE (ICLE-FR(+25); top part) and NE (LOCNESS; bottom part).

Collexemes that are strongly distinctive of FrLE include several verbs of change of state or location, including change, come, move, evolve and go away, as illustrated in (13).

(13)

The idea is not to point out that things were done wrong but to prove that we – human beings – are the ones who decide and who can make things change by educating people correctly. (ICLE-FR+25-18326)

Among the collexemes strongly distinctive of NE are two verbs (mostly) used as copulas, namely appear (LOR = 24.64) and sound (LOR = 22.83). By contrast, the copular verb be is strongly distinctive of FrLE (LOR = 23.49). Interestingly, be is redundant when used in a causative construction in ICLE-FR(+25). Thus, (14) could be more economically expressed as V. Woolf wants to make her reader aware of that. All of learners’ uses of be in [X make Y V_inf] could be replaced by adjectival causative constructions (as in [14]) or, in one case, by a past participle causative construction (15).

(14)

V. Woolf wants to make her reader be aware of that and of the fact that how hard he may work for peace and harmony, man will never reach it because war will always be there to destroy what he creates. (ICLE-FR-UC2001)

(15)

The purpose of this movement was at first to make women be recognised for what they really were. (ICLE-FR+25-16023)

5.2.2 Comparison of French learner English and general learner English

The results of the distinctive collexeme analysis of the [V_inf] slot of [X make Y V_inf] in FrLE and GenLE are shown in Figure 6 (see Appendix 5 for the full list of results). The verbs above the horizontal dashed line are more distinctive of FrLE, whereas those below the line are more distinctive of GenLE. Their position along the x-axis shows their (logged co-occurrence) frequency in the construction.

Figure 6:

Log odds ratio and logged co-occurrence frequency of the distinctive collexemes of [X make Y V_inf] in FrLE (ICLE-FR(+25); top part) and GenLE (ICLE without ICLE-FR; bottom part).

Many of the collexemes that were only slightly distinctive of NE or FrLE (Figure 5) are also only slightly distinctive of FrLE or GenLE (Figure 6), e.g. feel, think, believe, become or realize. These are collexemes that do not show a strong preference for one variety or the other.

Among the collexemes that were more clearly distinctive of FrLE than of NE, several verbs describing a change of state or location are also more distinctive of FrLE than of GenLE, e.g. go away, evolve, change and move. However, it is important to point out that we also find some verbs expressing a change of location among the collexemes distinctive of GenLE, such as go round, go and leave.

As for copular verbs, which were more distinctive of NE than of FrLE, some are also more distinctive of GenLE than of FrLE. This is the case of appear and sound. The copula be, although technically also more distinctive of GenLE, has such a low LOR (0.05) that it can be said to be common to French-speaking learners and learners from other mother tongue backgrounds. The verb is actually found in 13 out of the 16 components of ICLE, which suggests that it is a feature typical of learner English, regardless of learners’ mother tongue backgrounds. While in a few cases, the use of the verb is perfectly acceptable, as when it is combined with other infinitives, e.g. (16), most of the time the verb is redundant and the verbal causative construction could more economically be replaced by an adjectival or nominal causative construction, e.g. (17) or (18). This use could be related to the principle of redundancy, which Schneider (2012) claims is a general principle of human cognition likely to influence the process of second language acquisition (and hence is expected to potentially affect all learner varieties).

(16)

Almost every magazine or newspaper contains articles that are very optimistic in their nature, that make people laugh or just be happy. (ICLE-PO-PZ1048)

(17)

You gave us talents, dreams, emotions everything to make us be creative and imaginative. (ICLE-TS-NO1492)

(18)

And I think that it makes us to be adults. (ICLE-JP-WA3020)

6 Bringing CA and CIA together

In an attempt to identify potential cases of negative transfer from French in French-speaking learners’ collostructional preferences, the results of the CA (Section 4) and the CIA (Section 5) are now brought together. This combined approach relies on the above findings, which focus on verb classes, but also on Tables 3 and 4. These tables list the ten most distinctive collexemes of [X make Y V_inf] in FrLE and NE, respectively, and, when relevant, establish a correspondence with (1) the results of the CIA comparing FrLE and GenLE and (2) the results of the translation-based distinctive collexeme analysis comparing English (EN) and French (FR). The verbs in bold in the tables are those for which all analyses point in the same direction, i.e. showing the same distinctive (dis)preferences in French and FrLE.

In terms of verb classes, verbs expressing a change of state or location tend to be (i) more distinctive of French than of English, (ii) more distinctive of FrLE than of NE and (iii) more distinctive of FrLE than of GenLE. Although the exact verbs are not always the same and some individual verbs do not follow the general tendency (e.g. go and leave are more distinctive of GenLE than of FrLE), it seems as if this verb class might be transferred from French [X faire V_inf Y] to English [X make Y V_inf]. This is illustrated by (19), a French causative construction with the verb évoluer ‘evolve’, and (20), an English causative construction with the verb evolve in FrLE.

Table 3 includes a verb from this class, namely move, which fulfils all the above criteria (i–iii), but whose LOR for the French preference in the CA is extremely low (0.02). The verb change in the table is more distinctive of FrLE than of NE (ii) and GenLE (iii), but it is more associated with [X make Y V_inf] than with [X faire V_inf Y] (though with a low LOR of 0.67).

Interestingly, verbs expressing a change of state or location tend to be more typical of the [X cause Y V_to-inf] construction in English (Gilquin 2006). This suggests that French-speaking learners may transfer such verbs from [X faire V_inf Y] to [X make Y V_inf] (probably because of the formal similarity between these constructions), without realizing that these verbs are more typical of another causative construction in English. Favouring [X make Y V_inf] over [X cause Y V_to-inf] is a tendency that can be observed among other learner populations too, as predicted by the fact that the former construction is more frequent than the latter in native English and also more visible in pedagogical materials (see Gilquin 2010). This could explain the presence of some verbs of change of state or location in GenLE, but would not necessarily exclude the possibility of transfer in the case of French-speaking learners, resulting in more such verbs being distinctive of FrLE.

In contrast to verbs expressing a change of state or location, copular verbs tend to be (iv) more distinctive of English than of French, (v) more distinctive of NE than of FrLE and (vi) more distinctive of GenLE than of FrLE, with the same caveats as above. This suggests that French-speaking learners are different in dispreferring copular verbs in [X make Y V_inf], which could be due to the fact that copular verbs are not attracted to [X faire V_inf Y]. It is thus arguably the lack of copular verbs in the French causative construction that is transferred to English. In some cases, the French equivalents of the English copular verbs may be quite uncommon themselves. Table 4 lists the verb sound and its French equivalent sonner, which fulfil the three criteria of dispreference (iv–vi). This verb is used only once in the French causative construction and not as a copula. The closest equivalent of copular sound in French is the use of sonner with an adverb, which is however quite rare, occurring only twice in the whole Scientext.

Table 4 includes another potential copular verb: appear. While it is also more distinctive of NE (v) and GenLE (vi), the CA shows that it is slightly more distinctive of French than of English (LOR = 0.85). However, as pointed out in Section 4.2, in French it is mostly used as a verb of appearance rather than a copula. In fact, paraître would be more likely as a copular verb than apparaître (in Scientext, apparaître followed by an adjective occurs 93 times, as against 255 for paraître followed by an adjective). It is still interesting that French-speaking learners of English do not seem to transfer the use of apparaître as a verb of appearance in the English causative construction, which may have to do with the polysemy of the verb.^[5]

One major exception to the above tendency for copular verbs to be dispreferred by French-speaking learners is be, which comes second in Table 3. As we saw earlier, it is used redundantly in several learner English varieties, including FrLE, and could be the result of a general cognitive principle of redundancy in second language acquisition. Such a principle has been shown to be at work in different non-native varieties of English, materializing for example in the use of double conjunctions (e.g. Although … but …; Mesthrie 2006: 139) or in the use of a preposition or particle whose meaning is already expressed by the accompanying verb (e.g. rise up; Gilquin 2015b: 106). In the present case, the use of be could be seen as a special case of redundancy, namely “a strategy of maximizing transparency, increasing the degree of explicitness of marking syntactic relationships” (Schneider 2012: 67). The verb be makes the relationship between the object and the object complement explicit (e.g. make us be creative), whereas this relationship is kept implicit in the more standard adjectival causative construction (cf. make us creative).

Among the verbs in Table 3 not yet discussed, forget, act, discover, learn and stop could potentially be cases of collostructional transfer, although some of the LORs are quite low. Examples (21) and (22) show parallel uses of découvrir/discover in French and FrLE.

Several collexemes in Table 4 lack a correspondence in the ICLE data and, even more often, in the French data. This indicates that some verbs may not be produced by French-speaking learners because their equivalents in French do not occur in causative constructions. In the case of agree and trust, this may be due to the fact that the French equivalents involve paraphrases (être/se mettre d’accord and avoir/faire confiance, respectively), which may not be easily combined with [X faire V_inf Y]. The verb confess is topic-related, with all the occurrences of this collexeme being found in one and the same essay in LOCNESS.

Finally, it should be underlined that, as suggested by the ICM and as known from the literature, transfer is by no means systematic. For example, although intervenir is the most distinctive collexeme for French in the contrastive distinctive collexeme analysis (see Figure 4), its equivalent in English, intervene, does not occur at all in [X  make Y V_inf] in ICLE-FR(+25).

7 Conclusion

Using the ICM, this article has sought to highlight possible cases of collostructional transfer in the [V_inf] slot of [X make Y V_inf] in FrLE. The different corpus-based comparisons suggest that French-speaking learners’ preference for verbs expressing a change of state or location and some individual verbs like discover could be cases of collostructional transfer from French to English. Copular verbs (except be), on the other hand, are arguably dispreferred by French-speaking learners as a result of these verbs not being typical of [X faire V_inf Y]. While the presence of transfer, an essentially cognitive phenomenon, cannot be ascertained on the basis of corpus data alone, the triangulation employed here provides converging evidence that transfer might be at work in some of the learners’ collostructional preferences, thus going beyond transfer-related claims that are based on one source of evidence only.

More generally, this article has shown how to combine contrastive collostructional analysis and contrastive interlanguage collostructional analysis to identify possible cases of collostructional transfer. Among the limitations of the study are the relatively small amount of data and the very low association measures of certain collexemes. Among the possible areas for improvement are the treatment of formally similar but semantically different collexemes such as appear/apparaître. A sense-based approach like that adopted in Gilquin (2013), using semantically tagged verbs as input for the collostructional analysis, would likely provide more refined results, which could be valuable to further characterize collostructional transfer.