The effect of coordinate structure on the licensing of cto-clauses in noun complement constructions: an experimental study

Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

Institute for Linguistic Studies RAS

The effect of coordinate structure on the licensing of cto-clauses in noun complement constructions: an experimental study

M. Knyazev

Previously, cto-clause complements of nouns were found to be more acceptable in constructions with collocations as opposed to non-collocations, which was taken to support the hypothesis whereby cto-clauses are introduced by a silent preposition licensed by incorporation into a complex predicate (created in collocational constructions). The paper presents the results of an acceptability rating study testing the prediction of this account, namely, that embedding the noun in a coordinate structure should block the incorporation and hence licensing of cto-clauses. In addition, an alternative hypothesis is examined according to which cto-clauses are more acceptable in collocations due to their higher frequency.

Key words: Russian, experimental syntax, noun complement clauses, collocation, coordination, frequency

М.Ю. Князев

Институт лингвистических исследований Российской академии наук

Экспериментальное исследование приемлемости придаточных с союзом что в функции сентенциального актанта существительного в конструкциях с сочинением

Предыдущие исследования показали, что придаточные со что при существительных более приемлемы в конструкциях с коллокациями, чем в конструкциях без коллокаций. Это может служить подтверждением ранее высказанной гипотезы о том, что такие придаточные вводятся нулевым предлогом, лицензируемым путем инкорпорации в сложный глагол в конструкциях с коллокациями. В настоящей работе приводятся результаты эксперимента на вынесение суждений о приемлемости, проверявшего предсказание данной гипотезы, согласно которому сочинительная конструкция будет блокировать инкорпорацию и тем самым приводить к неграмматичности придаточного. Также в эксперименте проверялась альтернативная гипотеза, согласно которой большая приемлемость придаточных при коллокации связана с большей частотностью последних.

Ключевые слова: русский язык, экспериментальный синтаксис, придаточные при существительных, коллокация, сочинение, частотность

Introduction

Like their English counterparts, declarative sentential complements in Russian (cto-clauses) can appear not only in direct (nominative subject / accusative object), but also in oblique/PP positions, where they alternate with the (P +) to + cto-clause construction (to,cto-clauses), which consists of the demonstrative to `that' case-marked by P or V and followed by a cto- clause, as shown in (1a).

(1) Ona uverena she.NQM certain

(v tom), cto

in it.Loc that

on pridet.

he will come

Previous work aimed to uncover semantic/grammatical factors governing the choice between the two clause types [Knyazev, 2016] found that there is a strong preference to realize complements of nominalizations like `hope', `conviction', etc. and of relational nouns such as `likelihood', etc. as to,cto- clauses (but not cto-clauses) unless the noun forms a `set collocation' with the higher verb (in which case both clause types are possible), cf. collocation `express conviction' in (2a) vs. non-collocation `strengthen (one's) conviction' in (2b). For further discussion of the contrast between cto- and to,c7o-clauses see [Kobozeva, 2013].

clause sentence actant

(2) a. Collocation

The contrast in (2), referred to as the `collocational restriction' (on cto- clause complements of nouns), was later tested in an acceptability rating study [Knyazev, to appear a] using a 2 x 2 factorial design crossing factors clause type and (non-)collocation, following experimental work on island effects [Sprouse et al., 2016]. In that experiment, which contrasted 12 collocations and non-collocations of the same N, as in (2), a (super-additive) interaction was found showing that, although cto-clauses were associated with a decrease in acceptability relative to to,cto-clauses in both constructions, this decrease was 0.43 points greater in the non-collocational constructions.

These results were taken to support the grammatical account of the collocational restriction proposed in [Knyazev, 2016], according to which:

a) to,cto-clauses in oblique/PP positions are introduced by a silent preposition, which must be licensed by abstract incorporation into a higher verb;

b) the noun can reanalyze with the verb to form a complex verb only in collocational constructions.

Given the logic of experimental syntax studies such as [Sprouse et al., 2016], the decrease in acceptability of cto-clauses in non-collocational constructions can be explained by a (grammatical) violation of the licensing conditions associated with silent P.

Although the experimental results are consistent with the proposed grammatical account, they are not conclusive as the collocations and noncollocations (within a sentence set) in the experiment above differed widely along various dimensions including broad semantics and frequency. For example, one cannot exclude an alternative hypothesis according to which cto-clauses are simply stored in memory along with associated V-N combinations in an item-based fashion, whereas to,cto-clauses are fully productive and depend only on the semantics of the construction (henceforth, the “frequency hypothesis”). The frequency hypothesis would predict a contrast in acceptability for cto-clauses between collocational V-N combinations, which are more frequent, and non-collocational ones, which are less frequent, but no contrast for to,cto-clauses. This difference would result in a super-additive interaction.

This hypothesis was tested in [Knyazev, to appear b], which contrasted 12 pairs of semantically similar collocations with a higher vs. lower corpus frequency (e.g., vyrazit' uverennost' `express conviction' vs. vyskazat' uverennost' `voice hope') using a forced-choice task. Although the results showed no effect of frequency and hence no support for a frequency-based account, the difference in the design of this experiment from the design of the experiment in [Knyazev, to appear a], makes it hard to compare the two experiments directly. Moreover, in a follow-up analysis a marginal correlation was found between the proportion of cto-clause responses and the log frequency of V-N, suggesting that frequency might still play some role in acceptability.

In view of the limitations of the studies reported above, a more direct way of testing the predictions of the account in [Knyazev, 2016] is called for. In particular, the account predicts that if incorporation of silent P (by hypothesis, introducing a cto-clause) is blocked by some syntactic configuration such as embedding in a coordinate structure, a cto-clause will become ungrammatical. In addition, a frequency-based account must be tested within the same design in order to be able to directly compare the two effects. This paper reports the results of an acceptability rating study which (in a single design) simultaneously tested the effects of embedding the noun in a coordinate structure and the effect of frequency of the V-N collocation on the acceptability of cto-clauses.

The paper is structured as follows. In section 2 the main competing hypotheses are introduced, and the design of the experiment is discussed. Section 3 discusses the materials and the procedure used in the experiment. Section 4 presents and discusses the results of the experiment. Section 5 concludes.

The hypotheses and design

The coordination hypothesis

According to [Knyazev, 2016], cto-clause complements are underlyingly nominal expressions and are subject to the Case Filter. The crucial assumption of the account is that in apparent `Caseless' positions, e.g., those associated with oblique/PP selecting verbs, cto-clause are licensed by the silent preposition P_HOLD, which has the semantics of the relation of holding propositional content. It is further assumed that P_HOLD must incorporate into a verbal head in order to get licensed (the incorporation has the semantic effect of predicate conjunction). Assuming that in collocational V-Ns such as (2a), N itself can abstractly incorporate into V to create a complex predicate (cf. the analysis of make the claim in [Davies, Dubinsky, 2003]), P_HOLD associated with such V-Ns will be licensed by virtue of incorporation into this complex predicate, as schematized in (3a). According to standard assumptions, this process must proceed cyclically. First, P_№ld incorporates into N and then the [Phold + N] complex incorporates into V. By contrast, in non-collocational V-N, as in (2b), no complex predicate will be created and thus P_HOLD and the cto-clause will fail to be licensed.

This account predicts that even in collocational V-N such as (2a) whenever incorporation of N to V is disrupted, a complex predicate will fail to be created, thus blocking licensing of the cto-clause. One such configuration is coordinate structure, generally taken to be an island for movement (including incorporation). The prediction then is that if the complementtaking N is conjoined with a noun phrase (xNP), the incorporation of this N (or, more precisely, the P_HOLD + N complex) into V will be blocked, leading to ungrammaticality, as in (3b).

(3) a. [_v, [[P_H0LD + N] + V] [_Np [P_H0LD + N] [pp P_H0LD Lp. cto... ]]] (-(2a))

b. *^[v, ^{V [xNP} & [pp ^[PHOLD ^{+ N] [}PP ^PHOLD ^[xNP ^&t0 - ^]]]] t X I(-(4))

To test this hypothesis, examples like (4), with coordination, were contrasted with examples like (2a) above, without coordination. Given the logic of the previous studies [Knyazev, to appear a], the prediction is that in examples like (2a), cto-clauses will lead to a greater decrease in acceptability (relative to to,cto-clauses) than in examples like (4), which is equivalent to a non-zero (positive) difference between the differences between cto and to, cto in the two conditions, so-called “difference-in-differences”, or DD, which can be viewed as the measure of the “coordination effect”, as shown in (5), see [Sprouse et al., 2016]. In statistical terms, this hypothesis predicts a (super-additive) interaction between the factors ±coordination and CLAUSE TYPE (to, Cto vs. Cto).

(4) Ona vyrazila [javnoe nedoumenie] i [uverennost'

she.NOM expressed obvious puzzlement and conviction.ACc

^??(v tom), cto on pridet]].

in it.LOc that he will come

(5) Coord_to._o - Coord._o > Non-Coord _to ^_o - Non-Coord_Яо *»

DD - (Coord _toJto - Coord _ao) - (Non-Coord _to._o - Non-Coord._o) > 0

The frequency hypothesis

According to the frequency hypothesis, discussed in [Knyazev, to appear b], the contrast between examples with collocational V-N in (2a) and non-collocational V-N in (2b) follows from the higher token frequency of the former. In accordance with usage-based approaches, the hypothesis assumes that experience (operationalized as token frequency) plays an important role in the acquisition of and mature competence with particular constructions. Crucially, it is further assumed that the effect of token frequency is modulated by higher “regularity” of the construction (operationalized as a higher type frequency) so that more regular constructions are less affected by token frequency (as compared to construction with a lower type frequency) as they are less dependent on experience. This leads to what has been described as Regularity x Frequency interactions [Christiansen, Chater, 2016]. Assuming that cto-clauses in oblique/PP positions are less regular than to,cto-clauses, they will be more strongly affected by the higher frequency of V-N (“frequency effect”), as shown in (6), accounting for the contrast between collocational and non-collocational V-N in (2a)-(2b).

(6) MoreFreq_c7o - LessFreq_c7o > MoreFreq_/o._o - LessFreq_/o._o *»

DD = (LessFreq_/o._o - LessFreq^) - (MoreFreq_/o._o - MoreFreq^) > 0

The frequency hypothesis in (6) further predicts that the same contrast will be obtained between collocational V-N with higher vs. lower token frequencies (in construction with cto-clauses). To test this hypothesis, collocations like (2a) were contrasted with semantically and structurally identical collocations with lower frequency, as shown in (7). In statistical terms, this hypothesis predicts a (super-additive) interaction between factors FREQUENCY and CLAUSE TYPE.

(7) Ona vyskazala uverennost' (v tom),

she.NOM voiced conviction.ACC in it.LOC

cto on pridet.

that he will come

Note that the contrast between (2a) and (7), should it occur, is not predicted by the grammatical account in (3), as both types of V-Ns are presumably derived by incorporation. Similarly, the contrast between (2a) and (4) is not predicted by the frequency hypothesis (at least without modification) as they involve the same surface strings “V + N + cto”. Thus, the results of the experiment will allow to decide which hypothesis is more consistent with the data.

Materials and procedure

The experimental items consisted of twelve sets of six lexically matched sentences given in Table 1. Each set included a pair of V-N collocations which were semantically and structurally similar but differed in frequency as well as one coordinated collocation based on the more frequent V-N, each with a cto- and a to,cto-clause. The pairs of collocations were selected based on a prior corpus study in such a way that the higher frequency V-Ns had a higher token frequency of cto-clauses, see Table 2. The corpus study was based on the texts written after 1950 in the Russian National Corpus (RNC), ruscorpora.ru. See [Knyazev, to appear b] for details. The coordinated collocations were constructed by (left-)conjoining the noun in the collocation by some semantically appropriate noun modified by an adjective. In order to exclude an analysis with coordination between N-heads, nouns that disallow a cto-clause in this construction were chosen as the first N. Additionally, the two Ns typically differed in gender and/or most adjectives were semantically incompatible with the second N. The actual sentences that participants rated were constructed by slightly modifying naturally occurring examples found on the Web. Experimental sentences based on nine pairs of high vs. lower frequency V-Ns in Table 1 (sets 1-9) were directly taken from [Knyazev, to appear b].

The experimental sentences were divided into six lists in a Latin square design and interspersed with 24 fillers with different syntactic structures (8 grammatical, 8 ungrammatical and 8 of intermediate acceptability). Participants were instructed to rate the naturalness of experimental sentences on a 7-point scale (1 indicating a totally unnatural-sounding sentence). The experiment was hosted on Google Forms and advertised via social media. One hundred and seventy-five people participated in the experiment; their mean age was 28.05 (range: 16-62).

Table 1

Materials used in the experiment

Table 2

Frequencies of cto- and a to,cto-clauses with more and less frequent collocations (RNC)

Results and discussion

Overall results

The ratings were z-score transformed in accordance with the usual practice in experimental syntax [Sprouse et al., 2016]. Grammatical fillers received the (transformed) rating of 0.61 (SD = 0.66), ungrammatical fillers the rating of -1.12 (SD = 0.51) and fillers of intermediate acceptability the rating of -0.4 (SD = 0.88). The (transformed) condition means are given in Table 3.

Table 3 Z-score means (SD) in the experiment

The ratings were entered into a linear mixed-effects model with clause type, construction type and their interaction as fixed effects. Construction type was treatment coded with more frequent/non-coordinated V-N as the baseline contrasted with less frequent and coordinated V-N. Following [Barr et al., 2013], a maximal random effect structure that allowed convergence was used.

The model included random by-subject slopes for clause type and construction type, random item intercept and random by-item slopes for clause type and construction type and their interaction. P-values were obtained using Satterthwaite approximation from the lmerTest package for R.

The model revealed a main effect of clause type such that sentences with cto-clauses were rated lower than sentences with to,cto-clauses (Estimate = -0.35, SE = 0.07, t = -5.10, p < 0.001). The model also revealed a main effect of construction type such that sentences with coordination were rated lower than sentences with non-coordinated/more frequent collocations. (Estimate = -0.25, SE = 0.06, t = -3.48, p < 0.004). Interestingly, no interaction between clause type and construction type was observed as shown by the fact that the coefficients for both interaction terms in the model output were not significantly different from zero (Сто * LessFreq: Estimate = 0.06, SE = 0.06, t = 0.89, p = 0.39; Сто * Coord: Estimate = -0.11, SE = 0.08, t = -0.38, p = 0.19). These results are plotted in Fig. 1.

Fig. 1. Z-score means (SE) in the experiment

Results by items: Coordination

Although the interaction between coordination and clause type was not significant, we see a numerical trend in the predicted direction, i.e., cto-clauses were associated with a 0.11 points greater decrease in acceptability in the coordination condition. Thus, it was decided to inspect individual sets, as shown in Table 4.

Four out of 12 items (sets 1, 6, 8 and 12), shaded in Table 4, showed a positive DD_coord, as defined in (5), with range 0.33-0.48, which is above the minimal threshold for island effects (= 0.25) in [Kush et al., 2018]. Since there is no standardly accepted threshold for a grammatical (interaction) effect [Sprouse et al., 2016], the threshold suggested in [Kush et al., 2018] was used. In addition, set 7 showed a DD_coord close to this threshold (0.20). One interesting feature that these sets share is that they are precisely those that showed the smallest effect of cto-clause (-0.02-0.25), shown in bold, operationalized as the difference between to,cto- and cto-clauses in the baseline condition (Cto.Effect).

This observation was confirmed by a strong negative correlation between DD_coord and the effect of cto-clause (r = -0.89, p < 0.001), suggesting that a stronger decrease shown by cto-clauses in the baseline condition is associated with a weaker effect of coordination. By contrast, there was no correlation between DD_coord and the (independent) effect of coordination (Coord.Effect), operationalized as the difference between coordinated and non-coordinated collocations in the to,cto condition (r = 0.12, p = 0.69). At first glance, the negative correlation between DD_coord and the effect of cto- clause is expected given that DDcoord is calculated by substracting the effect of cto-clause (in the baseline condition) from the effect of cto-clause in the coordinated condition, cf. (5). Note, however, that this is only the case if the latter effect is constant, which need not be the case. Note also that DDcoord is also equivalent to the difference between the effect of coordination in the cto- condition and the effect of coordination (in the to, cto condition), i.e., Coord.Effect, as in (i). Yet, there was no correlation between DDc₀₀rd and Coord.Effect.

Table 4 Experimental effects by items

Results by items: Frequency

By-item DD-scores for the interaction between frequency and clause type (DD_freq), cf. (7), were inspected in a similar way. In this case we see little evidence for any interactions: only set 6 showed an effect greater than 0.25 (0.40); in addition, two others (sets 7 and 12) showed an effect of 0.20-0.23. There does not seem to be any clear property that sets these sets apart. Yet, we also see a marginally significant negative correlation between the effect of cto-clause and DD_freq (r = -0.59, p = 0.05), suggesting that the frequency effect is somewhat stronger for those sets that have a weaker effect of cto- clause in the baseline (more frequent) condition.

Since frequency was treated as a categorical variable, the difference in frequency between particular pairs of collocations in the experiment might have been too small to lead to a visible effect. Therefore, it was decided to test whether there is association between the log frequency of a collocation in construction with a cto-clause (see Table 2), and the effect of cto-clause for all 24 collocations (union of Cto.MoreFreq and Cto.LessFreq), ignoring the fact the collocations were matched in the experiment. This correlation was not significant (r = -0.33, p = 0.11). However, the (negative) correlation between the proportion of cto-clause and the effect of cto-clause did reach significance (r = -0.44, p = 0.03), suggesting that collocations with a higher proportion of cto-clauses tend be less effected by the decrease in acceptability associated with cto-clauses.

Discussion

Overall, the experiment did not provide evidence for the effect of coordination (operationalized as a super-additive interaction between coordination and clause type). While there were independent lowering effects of both cto-clause and coordination, there was no decrease in acceptability of cto-clauses with coordination above and beyond those effects (i.e., we see a linear additive effect). Thus, the results fail to provide direct support for the grammatical account in (3). Nonetheless, 4 to 5 out of 12 items were associated with a stronger decrease in acceptability of cto-clauses in the coordination condition (0.33-0.48), as compared to the baseline. In addition, all items showed a stable lowering effect of cto-clause in the coordination condition (0.33-0.70), see column Cto.Coord in Table 4. Given the operationalization of the coordination effect in (5), it is possible that this effect was “artificially” reduced by an independent effect of cto-clause. This interpretation is supported by the fact that the coordination effect is seen precisely in those V-Ns that do not show an effect of cto-clause, suggesting that it is somehow “absorbed” by the effect of cto- clause, see [Hofmeister et al., 2014] for some discussion. This raises questions about how the coordination effect should be operationalized. Perhaps it should be defined factorially (see [Sprouse et al., 2016]) only when other effects are not simply controlled for but cancelled (i.e., when there is no independent effect of cto-clause). Or perhaps it should be defined in relative rather than factorial terms (i.e., as the effect of cto-clause in the coordinated condition). These questions are left for future work.

As for the frequency hypothesis, the results provide little evidence for it. Apart from the lack of interaction between frequency and clause type, only 1 to 3 items out of 12 (with no clear pattern) showed any effect. Moreover, the effect of cto-clause was not consistently observed in the less frequent condition (see column Cto.LessFreq in Table 4). While this might be due to a particular choice of materials so that a higher contrast in frequency might lead to a stronger effect, on the whole, frequency remains an unlikely source of the previously reported super-additive interaction, in accord with [Knyazev, to appear b].

Conclusion

I have examined two potential explanations for the so-called collocational restriction, according to which cto-clause complements of nouns are associated with a higher decrease in acceptability (relative to to,cto-clauses) if the noun belongs to a V-N collocation [Knyazev, to appear a]. Namely, (a) the grammatical account, whereby cto-clauses are introduced by a null preposition licensed by incorporation into a complex verbal head created in collocational constructions; and (b) the frequency account, according to which collocations have a higher frequency, affecting the acceptability of cto-clauses (but not to,cto-clauses). The study focused on the specific prediction of the grammatical account, according to which coordination should block incorporation of the null preposition and thus “unlicense” cto-clauses. These two hypotheses were tested in an acceptability study with a factorial design crossing factors construction type (more frequent/ non-coordinated vs. less frequent vs. coordinated collocations) and clause type (to,cto- vs. cto-clauses), where both hypotheses were operationalized as a super-additive interaction as in [Sprouse et al., 2016]. While the results did not provide direct support for either account, there is indirect evidence that the coordination effect was obscured by an independent effect of cto- clause, suggesting that coordination effect might be restricted to those collocations that are equally acceptable with cto- and to,cto-clauses.

References

Barr et al., 2013 - Barr D.J., Levy R., Scheepers C., Tily H.J. Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language. 2013. Vol. 68. No. 3. Pp. 255-278.

Christiansen, Chater, 2016 - Christiansen M.H., Chater N. Creating Language. Cambridge, MA, 2016.

Davies, Dubinsky, 2003 - Davies W., Dubinsky S. On extraction from NPs. Natural Language & Linguistic Theory. 2003. Vol. 21. No 1. Pp. 1-37.

Hofmeister et al., 2014 - Hofmeister P., Staum Casasanto L., Sag I.A. Processing effects in linguistic judgment data: (super-)additivity and reading span scores. Language and Cognition. 2014. Vol. 6. No. 1. Pp. 111-145.

Knyazev, 2016 - Knyazev M. Licensing clausal complements: The case of cto- clauses. PhD diss., Utrecht University. 2016.

Knyazev, to appear a - Knyazev M. An experimental study of the distributional restriction on Russian cto-clause complements of nouns. Proceedings of FASL 26. To appear.

Knyazev, to appear b - Knyazev M. Frequency as a (non)-predictor of acceptability: Russian V-N collocations with cto-clause complements. Proceedings of FASL 27. To appear.

Kush et al., 2018 - Kush D., Lohndal T., Sprouse J. Investigating variation in island effects: A case study of Norwegian. Natural Language & Linguistic Theory.

Kobozeva, 2013 - Кобозева И.М. Условия употребления «то» перед придаточным изъяснительным с союзом «что» // Du mot au texte. Etudes slavo-romanes / Под ред. О. Инковой. Bern, 2013. С. 131-150. [Kobozeva I.M. Terms of use to before the subjunctive with the conjunction cto. Du mot au texte. ftudes slavo- romanes. O. Inkova (ed.). Bern, 2013. Pp. 131-150. (In Russ.)]

Sprouse et al., 2016 - Sprouse J., Caponigro I., Greco C., Cecchetto C. Experimental syntax and the variation of island effects in English and Italian. Natural Language & Linguistic Theory. 2016. Vol. 34. No. 1. Pp. 307-344.

Об авторе / About the author

Князев Михаил Юрьевич - PhD (лингвистика); научный сотрудник Отдела теории грамматики, Институт лингвистических исследований Российской академии наук, г. Санкт-Петербург

Mikhail Yu. Knyazev - PhD in Linguistics; Researcher at the Department of Grammar Theory, Institute for Linguistic Studies RAS, St. Petersburg, Russian Federation

Размещено на Allbest.ru