University of Groningen
Experimental investigations into the semantics of distributive marking
Bosnić, Ana
DOI:
10.33612/diss.171644158
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2021
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Bosnić, A. (2021). Experimental investigations into the semantics of distributive marking: Data from Serbian, Korean and Dutch. University of Groningen. https://doi.org/10.33612/diss.171644158
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Acquisition path of distributive markers
in Serbian and Dutch – evidence from an
act-out task
This chapter was published in a shorter form as:Bosnić, A. & Spenader, J. (2019). Acquisition path of distributive markers in Serbian and Dutch: Evidence from an act-out task. In Brown, M. M. & Dailey, B (eds.). BUCLD 43: Proceedings of the 43rd annual Boston University Conference on Language Development. Vol.1, 94-108. Cascadilla Press.
Acquisition path of distributive markers
in Serbian and Dutch – evidence from an
act-out task
This chapter was published in a shorter form as:Bosnić, A. & Spenader, J. (2019). Acquisition path of distributive markers in Serbian and Dutch: Evidence from an act-out task. In Brown, M. M. & Dailey, B (eds.). BUCLD 43: Proceedings of the 43rd annual Boston University Conference on Language Development. Vol.1, 94-108. Cascadilla Press.
2
Abstract
Adnominal distributive markers can be classified as distributive key (Dist-Key) markers, i.e., markers that attach to the argument that is being dis-tributed over, and distributive share (DistShare) markers, i.e., markers that attach to the argument that is being distributed. We tested the acquisition of distributivity in Serbian, a language that has both DistKey (svaki) and DistShare (po) markers, and compared it directly to Dutch, a language that only has DistKey (elke) markers. To do this we used an act-out task and we were able to elicit a range of interpretations that children have for sentences with and without distributive markers. We identified three major response types and found that Serbian children are significantly late in acquiring both DistKey and DistShare markers compared to Dutch children. In addition, since there is a third distributive competitor in Serbian (co-occurrence of the two distributive markers, svaki and po) we speculate it may affect the acquisition of DistKey and DistShare markers. Finally, we also saw that certain pragmatic factors seem to affect Dutch and Serbian children (and adults) differently, possibly resulting in the discrepancy between adult-like and non-adult-like responses within and between the two languages.
1 Introduction
This chapter investigates the acquisition path of the DistShare markers and their possible interpretations. Languages can be roughly divided into languages that have both DistKey and DistShare markers (e.g., Serbian, German, Korean, Japanese, etc.) and others that only have DistKey markers (e.g., Spanish, French, English, Dutch, etc.). While the DistKey markers have been intensively investigated in the past few decades, the acquisition of DistShare markers is still a relatively novel topic. What is interesting about the DistShare markers is that, apart from distributive readings over individuals, they also yield event-distributive readings, which is not the case with DistKey markers. A research question that then presents itself is how this affects acquisition. Could it be that the availability of additional distributive markers in Serbian, together with added complexity of differ-ent readings with DistShare markers, potdiffer-entially slow down the acquisition of one or both markers? We explore this question below.
The two types of markers differ syntactically in whether the marker attaches to an argument associated with the restrictor set (also called DistKey) or to an argument or a numeral associated with the scope of the sentence (also called DistShare). To illustrate, consider the sentences in (1), (2) and (3):
(1) The children are carrying a present. (2) [Each child] is carrying a present.
(3) Deca nose [po jedan poklon]. children.nom carry.pl distr one present.acc ‘Children are carrying one present each’.1
Sentences that are not marked with distributive markers such as (1), yield both distributive (Fig 1a) and collective (Fig 1b) readings, while distrib-utively marked sentences in English (2) and Serbian (3) tend to block the collective reading and only allow the distributive reading (Fig 1a).
a: Distributive reading b: Collective reading
Figure 1: Two dominant readings (distributive and collective) that arise with the sentence
in (1).
1 Note that po is not a binominal each, despite the similarities. Binominal each, although it syntactically attaches to the DistShare, it still is semantically associated with the DistKey. Furthermore, po, unlike binominal each, allows for other (event-related) readings. We omit a discussion of binominal each for now, but analyses of binominal each as a type of DistShare marker will be briefly addressed in Chapter 3.
2
Abstract
Adnominal distributive markers can be classified as distributive key (Dist-Key) markers, i.e., markers that attach to the argument that is being dis-tributed over, and distributive share (DistShare) markers, i.e., markers that attach to the argument that is being distributed. We tested the acquisition of distributivity in Serbian, a language that has both DistKey (svaki) and DistShare (po) markers, and compared it directly to Dutch, a language that only has DistKey (elke) markers. To do this we used an act-out task and we were able to elicit a range of interpretations that children have for sentences with and without distributive markers. We identified three major response types and found that Serbian children are significantly late in acquiring both DistKey and DistShare markers compared to Dutch children. In addition, since there is a third distributive competitor in Serbian (co-occurrence of the two distributive markers, svaki and po) we speculate it may affect the acquisition of DistKey and DistShare markers. Finally, we also saw that certain pragmatic factors seem to affect Dutch and Serbian children (and adults) differently, possibly resulting in the discrepancy between adult-like and non-adult-like responses within and between the two languages.
1 Introduction
This chapter investigates the acquisition path of the DistShare markers and their possible interpretations. Languages can be roughly divided into languages that have both DistKey and DistShare markers (e.g., Serbian, German, Korean, Japanese, etc.) and others that only have DistKey markers (e.g., Spanish, French, English, Dutch, etc.). While the DistKey markers have been intensively investigated in the past few decades, the acquisition of DistShare markers is still a relatively novel topic. What is interesting about the DistShare markers is that, apart from distributive readings over individuals, they also yield event-distributive readings, which is not the case with DistKey markers. A research question that then presents itself is how this affects acquisition. Could it be that the availability of additional distributive markers in Serbian, together with added complexity of differ-ent readings with DistShare markers, potdiffer-entially slow down the acquisition of one or both markers? We explore this question below.
The two types of markers differ syntactically in whether the marker attaches to an argument associated with the restrictor set (also called DistKey) or to an argument or a numeral associated with the scope of the sentence (also called DistShare). To illustrate, consider the sentences in (1), (2) and (3):
(1) The children are carrying a present. (2) [Each child] is carrying a present.
(3) Deca nose [po jedan poklon]. children.nom carry.pl distr one present.acc ‘Children are carrying one present each’.1
Sentences that are not marked with distributive markers such as (1), yield both distributive (Fig 1a) and collective (Fig 1b) readings, while distrib-utively marked sentences in English (2) and Serbian (3) tend to block the collective reading and only allow the distributive reading (Fig 1a).
a: Distributive reading b: Collective reading
Figure 1: Two dominant readings (distributive and collective) that arise with the sentence
in (1).
1 Note that po is not a binominal each, despite the similarities. Binominal each, although it syntactically attaches to the DistShare, it still is semantically associated with the DistKey. Furthermore, po, unlike binominal each, allows for other (event-related) readings. We omit a discussion of binominal each for now, but analyses of binominal each as a type of DistShare marker will be briefly addressed in Chapter 3.
2
In the study presented in this chapter, we restrict ourselves only to theindividual-distributive (atomic) interpretations of the DistShare markers in order to make the most direct comparison with Dutch. We are crucially interested in answering the following question: How does the acquisition path for individual distributivity differ for children learning a language with DistShare markers compared to children learning languages that only have DistKey markers?
The chapter is organized as follows – We first present the studies and most notable results concerning DistKey and DistShare markers (section 2.1). We introduce a new method, an act-out task, for testing comprehension of distributive markers and the underlying preferences for their inter-pretations in section 2.2. Section 3 presents the method in detail and we split the results based on the type of answer the children and adults gave (section 3.2.1 and 3.2.2). In section 4, we discuss different implications of the results, focusing on the adult-like responses being developed for Dutch and Serbian children (4.1), the role of pragmatics (4.2), and finally the late acquisition of DistKey and DistShare markers in Serbian (4.3). Finally, we summarize our observations that Serbian children may be significantly late in acquiring both DistKey and DistShare markers compared to their Dutch peers due to the presence of a DistShare marker, or even a third competitor that yields distributive readings (svaki+po) in section 5.
2 Background
2.1 Acquisition of distributive key and distributive share markers
Investigation on universal distributive (key) quantifiers has been vast – there is work on English distributive quantifiers each and/or every (Drozd 2001, Drozd et al. 2017, Lidz & Musolino 2002, Novogrodsky et al. 2013, Brooks & Braine 1996), spreading errors (Roeper et al. 2006, Roeper et al. 2011, Philip 1991, Brooks & Sekerina 2006, Sekerina et al. 2018), processing of quantifiers (Frazier et al. 1999, Sekerina & Sauermann 2015, 2017, Drozd et al. 2013), distributivity and numerals (Musolino 2004, 2009, Syrett & Musolino 2013, Syrett et al. 2012), and many others. However, most of these mainly focus on English, and studies of non-English dis-tributive quantifiers tend to compare their results to English quantifiers.
When it comes to children’s understanding of universal distributive (DistKey) quantifiers, it is important to highlight a number of things relevant to our study. Children generally prefer distributive readings from a young age, and for the quantifiers each and every (or elke and iedere in Dutch), research shows that children realize that the presence of these quantifiers results in distributive readings by the age of 6 (de Koster et al. 2017, Brooks & Braine 1996, Drozd 2001, Drozd et al. 2017). At the same time, there is evidence that children allow collective readings with distrib-utively quantified sentences at a higher rate than the adults. However, it is
also important to note that adults tested in English and Dutch sometimes accept collective scenarios with distributive quantifiers (see de Koster et al. 2017 and Rouweler & Hollebrandse 2015).
Another well-known problem in universal quantifier acquisition is the so-called overexhaustive and underexhaustive2 errors (Roeper et al. 2006;
2011, Brooks & Sekerina 2006). Although these errors are not a topic of our present study, it is worth noting that, although children seem to understand the distributive force of distributive quantifiers early in Dutch and English, their interpretation of these quantifiers is still not adult-like. Children continue making overexhaustive (also called spreading) errors when considerably older (even at the age of 11 in Dutch as observed in de Koster et al. (2018)). We thus see that certain properties of universal distributive quantifiers are still acquired later.
The acquisition of DistShare markers, on the other hand, is largely an unexplored territory, considering the large number of languages that have such markers. There is, however, work by Knežević (2015) or Knežević & Demirdache (2017; 2018) on the acquisition of the DistShare marker po in Serbian and there is also research on Hungarian on numeral reduplication which also has a distributive import (Kiss et al. 2013, Kiss & Zétényi 2018).3
In the following subsections we cover several important studies on nu-merically quantified expressions and distributive quantifiers in more detail in order to create a baseline for our own study that we present in section 3.
2.1.1 NQEs and scope
Let us start with the comparative data from two studies on two different languages – Musolino’s experiment (2009) on English number words and a replication of this study by Knežević’s (2015) done in Serbian. The important difference is, of course, that Serbian also has DistShare markers while English does not. We present the most relevant results (specifically NQEs) below.
Musolino (2009) tested how children understand scopal and non-scopal relationships (i.e., logical syntactic properties) between numerically quan-tified expressions (NQEs). He used a picture verification task (a variant of a truth value judgment task that includes the combinations of pictures and accompanying sentences) with 32 English speaking 5-year-olds and 32 adult speakers. The design included eight conditions consisting of two types of sentences (4):
2 Overexhaustive errors refer to the incorrect rejection of the situations described with universal quan-tifiers where there is not a 1-to-1 pairing of the agents and objects (e.g., if Figure 1a contained an extra present, some children may say no for the description Each child is holding a present, and explain that the extra present does not have a child). Underexhaustive errors are less common and disappear earlier and refer to the incorrect acceptance of a situation when there is an extra agent (e.g., an additional child without a present).
3 When it comes to DistShare markers such as po and spreading errors specifically, some data also exists on overexhaustive errors with Russian po and every (kazhdyj) (Sekerina & Sauermann 2010), where 5-year-old children show a significant percentage of overexhaustive errors, but this study remained unpublished.
2
In the study presented in this chapter, we restrict ourselves only to theindividual-distributive (atomic) interpretations of the DistShare markers in order to make the most direct comparison with Dutch. We are crucially interested in answering the following question: How does the acquisition path for individual distributivity differ for children learning a language with DistShare markers compared to children learning languages that only have DistKey markers?
The chapter is organized as follows – We first present the studies and most notable results concerning DistKey and DistShare markers (section 2.1). We introduce a new method, an act-out task, for testing comprehension
of distributive markers and the underlying preferences for their inter-pretations in section 2.2. Section 3 presents the method in detail and we split the results based on the type of answer the children and adults gave (section 3.2.1 and 3.2.2). In section 4, we discuss different implications of the results, focusing on the adult-like responses being developed for Dutch and Serbian children (4.1), the role of pragmatics (4.2), and finally the late acquisition of DistKey and DistShare markers in Serbian (4.3). Finally, we summarize our observations that Serbian children may be significantly late in acquiring both DistKey and DistShare markers compared to their Dutch peers due to the presence of a DistShare marker, or even a third competitor that yields distributive readings (svaki+po) in section 5.
2 Background
2.1 Acquisition of distributive key and distributive share markers
Investigation on universal distributive (key) quantifiers has been vast – there is work on English distributive quantifiers each and/or every (Drozd 2001, Drozd et al. 2017, Lidz & Musolino 2002, Novogrodsky et al. 2013, Brooks & Braine 1996), spreading errors (Roeper et al. 2006, Roeper et al. 2011, Philip 1991, Brooks & Sekerina 2006, Sekerina et al. 2018), processing of quantifiers (Frazier et al. 1999, Sekerina & Sauermann 2015, 2017, Drozd et al. 2013), distributivity and numerals (Musolino 2004, 2009, Syrett & Musolino 2013, Syrett et al. 2012), and many others. However, most of these mainly focus on English, and studies of non-English dis-tributive quantifiers tend to compare their results to English quantifiers.
When it comes to children’s understanding of universal distributive (DistKey) quantifiers, it is important to highlight a number of things relevant to our study. Children generally prefer distributive readings from a young age, and for the quantifiers each and every (or elke and iedere in Dutch), research shows that children realize that the presence of these quantifiers results in distributive readings by the age of 6 (de Koster et al. 2017, Brooks & Braine 1996, Drozd 2001, Drozd et al. 2017). At the same time, there is evidence that children allow collective readings with distrib-utively quantified sentences at a higher rate than the adults. However, it is
also important to note that adults tested in English and Dutch sometimes accept collective scenarios with distributive quantifiers (see de Koster et al. 2017 and Rouweler & Hollebrandse 2015).
Another well-known problem in universal quantifier acquisition is the so-called overexhaustive and underexhaustive2 errors (Roeper et al. 2006;
2011, Brooks & Sekerina 2006). Although these errors are not a topic of our present study, it is worth noting that, although children seem to understand the distributive force of distributive quantifiers early in Dutch and English, their interpretation of these quantifiers is still not adult-like. Children continue making overexhaustive (also called spreading) errors when considerably older (even at the age of 11 in Dutch as observed in de Koster et al. (2018)). We thus see that certain properties of universal distributive quantifiers are still acquired later.
The acquisition of DistShare markers, on the other hand, is largely an unexplored territory, considering the large number of languages that have such markers. There is, however, work by Knežević (2015) or Knežević & Demirdache (2017; 2018) on the acquisition of the DistShare marker po in Serbian and there is also research on Hungarian on numeral reduplication which also has a distributive import (Kiss et al. 2013, Kiss & Zétényi 2018).3
In the following subsections we cover several important studies on nu-merically quantified expressions and distributive quantifiers in more detail in order to create a baseline for our own study that we present in section 3.
2.1.1 NQEs and scope
Let us start with the comparative data from two studies on two different languages – Musolino’s experiment (2009) on English number words and a replication of this study by Knežević’s (2015) done in Serbian. The important difference is, of course, that Serbian also has DistShare markers while English does not. We present the most relevant results (specifically NQEs) below.
Musolino (2009) tested how children understand scopal and non-scopal relationships (i.e., logical syntactic properties) between numerically quan-tified expressions (NQEs). He used a picture verification task (a variant of a truth value judgment task that includes the combinations of pictures and accompanying sentences) with 32 English speaking 5-year-olds and 32 adult speakers. The design included eight conditions consisting of two types of sentences (4):
2 Overexhaustive errors refer to the incorrect rejection of the situations described with universal quan-tifiers where there is not a 1-to-1 pairing of the agents and objects (e.g., if Figure 1a contained an extra present, some children may say no for the description Each child is holding a present, and explain that the extra present does not have a child). Underexhaustive errors are less common and disappear earlier and refer to the incorrect acceptance of a situation when there is an extra agent (e.g., an additional child without a present).
3 When it comes to DistShare markers such as po and spreading errors specifically, some data also exists on overexhaustive errors with Russian po and every (kazhdyj) (Sekerina & Sauermann 2010), where 5-year-old children show a significant percentage of overexhaustive errors, but this study remained unpublished.
2
(4) (a) Two NQEs: Three boys are holding two balloons.
(b) Each in the object position (each-object): Three boys are holding each balloon.
The sentences were paired with four types of picture stimuli (see Figure 2 below, taken from Musolino (2009:31-32)). Knežević (2012, 2015) repli-cated this study with two NQEs and sentences with a Serbian quantifier
svaki in the object position with 20 adults and 56 children, divided in three
age groups (13 5-year-olds, 25 7-year-olds and 18 9-year-olds).
a: Subject-wide scope
(distributive) b: Object-wide scope c: Collective (each-all) d: Cumulative
Figure 2: Test pictures for scopal and non-scopal subject-object relationships, tested by
Musolino (2009).
Comparing the two languages, there seems to be a striking cross-linguis-tic difference between English and Serbian parcross-linguis-ticipants, and also between Serbian adults and children. See Table 1 below with the most relevant results with adults and Serbian and English 5-year-olds:
Table 1: A tabular overview of the most important results from an English (Musolino 2009)
and a Serbian (Knežević 2015) study for the sentences with two NQEs. Mean acceptance rates given in cells.
Test sentence Tri dečaka drže dva balona. / Three boys are holding two balloons.
Picture Distributive Collective
English adults 82.8 100
Serbian adults 15.0 92.5
English children (5yo) 78.1 98.4
Serbian children (5yo) 80.8 46.2
If we compare English and Serbian adults, we see that there is a clear difference in the acceptability of distributive readings with sentences with two NQEs – while English adults accept distributive readings (though less so than collective readings), Serbian adults seem to reject distributive readings almost completely (15%). At the same time, Serbian children differ from Serbian adults by accepting the distributive readings at a very high rate (and, in fact, they are indistinguishable from English children). When it comes to collective readings, we see that English adults and English children both find collective readings highly acceptable with NQEs, while Serbian adults and Serbian children clearly differ from each other, i.e.,
children seem to dislike collective readings with NQEs, while the opposite holds for adults. In other words, whereas English children seem to be adult-like in their interpretation of NQE sentences with both collective and distributive readings, accepting both, Serbian children have to learn that the distributive readings are not really acceptable, and that the collective readings are fine.4
Before moving to Serbian data with the DistShare marker po, there is another noteworthy study on NQEs in English worth mentioning. Namely, Syrett & Musolino (2013) did a series of experiments with English speaking children and adults with NQEs, testing both judgments and preferences. They confirmed that children and adults accept both collective and distrib-utive scenarios with ambiguous sentences (sentences with NQEs such as
Two boys pushed a car) significantly above chance. Distributive scenarios
were accepted less than collective ones, but there was no significant dif-ference between them. However, the more striking results are from the preference experiment – When the adults and children were asked to choose which situation better illustrates the ambiguous sentence Two boys
pushed a car, adults preferred collective scenarios (88.9%) to distributive
ones (11.1%), while children showed the opposite preference (31.5% for collective and 68.5% for distributive). Furthermore, in this study, English children’s non-adult performance is similar to the Serbian children in Knežević’s (2015) replication study of Musolino (2009) (even though these were different tasks) – children strongly prefer distributive readings with NQEs, showing a clear divergence from adult answers.
In order to continue building our baseline of how Serbian children seem to interpret sentences with NQEs once a distributive marker is added to the picture, we turn to a study in which Serbian children were tested with distributively marked sentences with the DistShare marker po and NQEs. These results then created a new puzzle in the acquisition of distributivity in Serbian – despite children being “distributive” with NQEs, they do not understand the distributive force of the DistShare marker po until very late. We discuss these findings in section 2.1.2 below.
4 Another cross-linguistic difference worth noting are the results for each-object/svaki-object sentences (e.g., Three boys are holding each balloon./Tri dečaka drže svaki balon.), reported in Knežević (2015). While English adults accept object-wide scope pictures (Figure 2b) 100% and collective pictures (Figure 2c) 85.9% of the time for each-object sentences, Serbian adults do not allow svaki to take wide scope over the numeral in the subject position and they only allow collective readings (72%) with svaki-object sentences. On the other hand, Serbian children accept all conditions above chance. Adult results seemingly support the claim that Serbian does not allow inverse scope readings. This discrepancy suggests that children may have difficulties determining scope, which makes them more permissive to different scopal readings, while adults clearly disallow inverse scope readings. This could be yet another property of Serbian quantifiers that Serbian children acquire late. This connects to a prevalent theoretical claim about inverse scope put forth by Progovac (1994) and Bošković (2012) that languages without articles (such as Japanese, Russian, Chinese and Serbian) do not allow inverse scope readings. However, this claim has not been experimentally tested and seems rather absolute, especially because in contexts in which, pragmatically, surface scope is impossible, e.g., Twin girls were born in every house – the inverse scope reading is available even in Serbian. In fact, we have confirmed this is a possible reading with several Serbian speakers. This larger topic goes beyond the scope of this chapter, and more extensive experimental work is needed to make concrete claims about inverse scope in article-less languages, particularly about the availability vs. preference of such readings.
2
(4) (a) Two NQEs: Three boys are holding two balloons.
(b) Each in the object position (each-object): Three boys are holding each balloon.
The sentences were paired with four types of picture stimuli (see Figure 2 below, taken from Musolino (2009:31-32)). Knežević (2012, 2015) repli-cated this study with two NQEs and sentences with a Serbian quantifier
svaki in the object position with 20 adults and 56 children, divided in three
age groups (13 5-year-olds, 25 7-year-olds and 18 9-year-olds).
a: Subject-wide scope
(distributive) b: Object-wide scope c: Collective (each-all) d: Cumulative
Figure 2: Test pictures for scopal and non-scopal subject-object relationships, tested by
Musolino (2009).
Comparing the two languages, there seems to be a striking cross-linguis-tic difference between English and Serbian parcross-linguis-ticipants, and also between Serbian adults and children. See Table 1 below with the most relevant results with adults and Serbian and English 5-year-olds:
Table 1: A tabular overview of the most important results from an English (Musolino 2009)
and a Serbian (Knežević 2015) study for the sentences with two NQEs. Mean acceptance rates given in cells.
Test sentence Tri dečaka drže dva balona. / Three boys are holding two balloons.
Picture Distributive Collective
English adults 82.8 100
Serbian adults 15.0 92.5
English children (5yo) 78.1 98.4
Serbian children (5yo) 80.8 46.2
If we compare English and Serbian adults, we see that there is a clear difference in the acceptability of distributive readings with sentences with two NQEs – while English adults accept distributive readings (though less so than collective readings), Serbian adults seem to reject distributive readings almost completely (15%). At the same time, Serbian children differ from Serbian adults by accepting the distributive readings at a very high rate (and, in fact, they are indistinguishable from English children). When it comes to collective readings, we see that English adults and English
children both find collective readings highly acceptable with NQEs, while Serbian adults and Serbian children clearly differ from each other, i.e.,
children seem to dislike collective readings with NQEs, while the opposite holds for adults. In other words, whereas English children seem to be adult-like in their interpretation of NQE sentences with both collective and distributive readings, accepting both, Serbian children have to learn that the distributive readings are not really acceptable, and that the collective readings are fine.4
Before moving to Serbian data with the DistShare marker po, there is another noteworthy study on NQEs in English worth mentioning. Namely, Syrett & Musolino (2013) did a series of experiments with English speaking children and adults with NQEs, testing both judgments and preferences. They confirmed that children and adults accept both collective and distrib-utive scenarios with ambiguous sentences (sentences with NQEs such as
Two boys pushed a car) significantly above chance. Distributive scenarios
were accepted less than collective ones, but there was no significant dif-ference between them. However, the more striking results are from the preference experiment – When the adults and children were asked to choose which situation better illustrates the ambiguous sentence Two boys
pushed a car, adults preferred collective scenarios (88.9%) to distributive
ones (11.1%), while children showed the opposite preference (31.5% for collective and 68.5% for distributive). Furthermore, in this study, English children’s non-adult performance is similar to the Serbian children in Knežević’s (2015) replication study of Musolino (2009) (even though these were different tasks) – children strongly prefer distributive readings with NQEs, showing a clear divergence from adult answers.
In order to continue building our baseline of how Serbian children seem to interpret sentences with NQEs once a distributive marker is added to the picture, we turn to a study in which Serbian children were tested with distributively marked sentences with the DistShare marker po and NQEs. These results then created a new puzzle in the acquisition of distributivity
in Serbian – despite children being “distributive” with NQEs, they do not understand the distributive force of the DistShare marker po until very late. We discuss these findings in section 2.1.2 below.
4 Another cross-linguistic difference worth noting are the results for each-object/svaki-object sentences (e.g., Three boys are holding each balloon./Tri dečaka drže svaki balon.), reported in Knežević (2015). While English adults accept object-wide scope pictures (Figure 2b) 100% and collective pictures (Figure 2c) 85.9% of the time for each-object sentences, Serbian adults do not allow svaki to take wide scope over the numeral in the subject position and they only allow collective readings (72%) with svaki-object sentences. On the other hand, Serbian children accept all conditions above chance. Adult results seemingly support the claim that Serbian does not allow inverse scope readings. This discrepancy suggests that children may have difficulties determining scope, which makes them more permissive to different scopal readings, while adults clearly disallow inverse scope readings. This could be yet another property of Serbian quantifiers that Serbian children acquire late. This connects to a prevalent theoretical claim about inverse scope put forth by Progovac (1994) and Bošković (2012) that languages without articles (such as Japanese, Russian, Chinese and Serbian) do not allow inverse scope readings. However, this claim has not been experimentally tested and seems rather absolute, especially because in contexts in which, pragmatically, surface scope is impossible, e.g., Twin girls were born in every house – the inverse scope reading is available even in Serbian. In fact, we have confirmed this is a possible reading with several Serbian speakers. This larger topic goes beyond the scope of this chapter, and more extensive experimental work is needed to make concrete claims about inverse scope in article-less languages, particularly about the availability vs. preference of such readings.
2
2.1.2 NQEs and po
In her thesis, Knežević (2015) also reports experiments where she tested interpretations of NQEs and a DistShare marker po. We should also note that apart from being a DistShare marker in Serbian, po also has other meanings – it is a locative preposition, a verbal prefix and it can attach to adverbs and adjectives (more on this in Chapters 3, 4 and 5). However, since the DistShare po is relevant to understanding the acquisition of collective and distributive interpretations, this was the focus of Knežević’s study. Spe-cifically, in a 3x3 study, she tested three types of sentences – NQE sentences (henceforth po-less sentences), po on the object and po on the subject:5
(5) a. Dve devojke peru brod. two girl.nom.pl wash.3.pl.ps boat.acc.sg
‘Two girls are washing a boat.’
b. Dve devojke peru po brod. two girl.nom.pl wash.3.pl.ps distr boat.acc.sg ‘Two girls are separately washing a boat.’
c. Po dve devojke peru brod. distr two girl.nom.pl wash.3.pl.ps boat.acc.sg ‘Two girls are washing a boat at separate/different locations.’
The second independent variable was a short video stimulus with a recorded voice describing the scene as it was playing, similar to Syrett & Musolino’s (2013) study. This variable had three levels – subject-distributive (subject-wide scope), event-distributive and collective. See Figure 3 that shows Knežević’s stimuli (2015:172-173) below:
a: Subject-distributive (SD) b: Event-distributive (ED) c: Collective (COLL)
Figure 3: Three test scenarios that are paired with the test sentences in (5).
74 children were tested, divided in three age groups: 5-, 7- and 9-year-olds, and 21 adult speakers. The full experiment consisted of three blocks of questions, each with 26 items. Each block was followed by a short break so the children would not get overwhelmed or bored with the experiment. Each block, however, was essentially a 1x3 design – sentence type was split between the blocks (so one block only had po-less sentences with three types of videos, for example).
We first present the predictions and the results of adult responses for ref-erence when discussing the children’s results. We omit po-subject sentences
5 Just like DistShare in other languages (Korean, Japanese, Quechua, etc.), po can also attach on both arguments in the sentence, yielding only event-distributive readings. This reading was tested in an experiment done by Knežević (2015), but it will not be covered in this chapter.
here since they are not relevant for the present study. See Table 2 for the most relevant results:
Table 2: Mean yes-responses for Serbian adults and children for the six conditions in
per-centages. Predicted responses are indicated in a separate column. Shaded cells indicate cases where the children’s responses resembled the adult responses.
(Subject)-Distributive Collective
Prediction Adults 5yo 7yo 9yo Prediction Adults 5yo 7yo 9yo
po-less NO 37.3 95.8 64.7 33.3 YES 99.2 93.8 98.7 95.3
po-object YES 93.7 97.8 73.3 58.7 NO 2.4 86.2 48.6 44.7
The most important observation here is that children are generally not giving like responses across the board. The only consistent adult-like responses are with the po-less sentences with collective scenarios, which they overwhelmingly accept. Moreover, the adult-like response for the po-object sentence in a subject-distributive scenario for 5-year-olds probably cannot be imputed to the understanding of po as a distributive marker because 7-year-olds and 9-year-olds are performing much worse. This result is either the consequence of a yes-bias response for young children, or, more likely, given the results from other languages, a general preference for distributive readings in very young children.
It is also essential to note that collective interpretations with po-sentences are not only dispreferred for adults, but completely incorrect – there cannot be a single boat being washed, since po requires a plurality (of entities or events), and a collective scenario does not show such plurality. Judging from the results in Table 2, children seem to accept distributive readings with po-less sentences (but not so much at the age of 9), and, at the same time, they are not sensitive to distributive force of po until quite late – they accept collective scenarios with po which shows they are not adult-like even at the age of 9. Even though the results for the po-subject sentences (5c) and event-distributive readings (Figure 3b) are not shown here, overall, the children had problems with them as well (they were only adult-like with event-distributive readings paired with a po-subject sentence). The results seem to be quite complex because we also see a lot of development over time, but without clear indications what could be going on. We can see this by summarizing the main conclusions: (i) very young children accept everything, which suggest that they like distributive readings but are completely insensitive to the distributive marker, and (ii) older children reject po-sentences incorrectly with distributive pictures but correctly with collective pictures, which either implies that they notice the presence of po but are still unsure what it does, or that they are contrasting their no-answers with yes-answers for the po-less sentences.
In order to understand why young children accept distributive readings in sentences with NQEs but then do not understand the distributive property of po, we turned to a proposal that this may be related to mor-phosyntactic cues, namely verbal agreement, that may prompt distributive
2
2.1.2 NQEs and po
In her thesis, Knežević (2015) also reports experiments where she tested interpretations of NQEs and a DistShare marker po. We should also note that apart from being a DistShare marker in Serbian, po also has other meanings – it is a locative preposition, a verbal prefix and it can attach to adverbs and adjectives (more on this in Chapters 3, 4 and 5). However, since the DistShare po is relevant to understanding the acquisition of collective and distributive interpretations, this was the focus of Knežević’s study. Spe-cifically, in a 3x3 study, she tested three types of sentences – NQE sentences (henceforth po-less sentences), po on the object and po on the subject:5
(5) a. Dve devojke peru brod. two girl.nom.pl wash.3.pl.ps boat.acc.sg
‘Two girls are washing a boat.’
b. Dve devojke peru po brod. two girl.nom.pl wash.3.pl.ps distr boat.acc.sg ‘Two girls are separately washing a boat.’
c. Po dve devojke peru brod. distr two girl.nom.pl wash.3.pl.ps boat.acc.sg ‘Two girls are washing a boat at separate/different locations.’
The second independent variable was a short video stimulus with a recorded voice describing the scene as it was playing, similar to Syrett & Musolino’s (2013) study. This variable had three levels – subject-distributive (subject-wide scope), event-distributive and collective. See Figure 3 that shows Knežević’s stimuli (2015:172-173) below:
a: Subject-distributive (SD) b: Event-distributive (ED) c: Collective (COLL)
Figure 3: Three test scenarios that are paired with the test sentences in (5).
74 children were tested, divided in three age groups: 5-, 7- and 9-year-olds, and 21 adult speakers. The full experiment consisted of three blocks of questions, each with 26 items. Each block was followed by a short break so the children would not get overwhelmed or bored with the experiment. Each block, however, was essentially a 1x3 design – sentence type was split between the blocks (so one block only had po-less sentences with three types of videos, for example).
We first present the predictions and the results of adult responses for ref-erence when discussing the children’s results. We omit po-subject sentences
5 Just like DistShare in other languages (Korean, Japanese, Quechua, etc.), po can also attach on both arguments in the sentence, yielding only event-distributive readings. This reading was tested in an experiment done by Knežević (2015), but it will not be covered in this chapter.
here since they are not relevant for the present study. See Table 2 for the most relevant results:
Table 2: Mean yes-responses for Serbian adults and children for the six conditions in
per-centages. Predicted responses are indicated in a separate column. Shaded cells indicate cases where the children’s responses resembled the adult responses.
(Subject)-Distributive Collective
Prediction Adults 5yo 7yo 9yo Prediction Adults 5yo 7yo 9yo
po-less NO 37.3 95.8 64.7 33.3 YES 99.2 93.8 98.7 95.3
po-object YES 93.7 97.8 73.3 58.7 NO 2.4 86.2 48.6 44.7
The most important observation here is that children are generally not giving like responses across the board. The only consistent adult-like responses are with the po-less sentences with collective scenarios, which they overwhelmingly accept. Moreover, the adult-like response for the po-object sentence in a subject-distributive scenario for 5-year-olds probably cannot be imputed to the understanding of po as a distributive marker because 7-year-olds and 9-year-olds are performing much worse. This result is either the consequence of a yes-bias response for young children, or, more likely, given the results from other languages, a general preference for distributive readings in very young children.
It is also essential to note that collective interpretations with po-sentences are not only dispreferred for adults, but completely incorrect – there cannot be a single boat being washed, since po requires a plurality (of entities or events), and a collective scenario does not show such plurality. Judging from the results in Table 2, children seem to accept distributive readings with po-less sentences (but not so much at the age of 9), and, at the same time, they are not sensitive to distributive force of po until quite late – they accept collective scenarios with po which shows they are not adult-like even at the age of 9. Even though the results for the po-subject sentences (5c) and event-distributive readings (Figure 3b) are not shown here, overall, the children had problems with them as well (they were only adult-like with event-distributive readings paired with a po-subject sentence). The results seem to be quite complex because we also see a lot of development over time, but without clear indications what could be going on. We can see this by summarizing the main conclusions: (i) very young children accept everything, which suggest that they like distributive readings but are completely insensitive to the distributive marker, and (ii) older children reject po-sentences incorrectly with distributive pictures but correctly with collective pictures, which either implies that they notice the presence of po but are still unsure what it does, or that they are contrasting their no-answers with yes-answers for the po-less sentences.
In order to understand why young children accept distributive readings in sentences with NQEs but then do not understand the distributive property of po, we turned to a proposal that this may be related to mor-phosyntactic cues, namely verbal agreement, that may prompt distributive
2
readings with NQEs. The potential influence of morphosyntactic cues ondistributivity acceptance rates has been investigated experimentally with adults and children in Bosnić (2016), and the results are presented in the following section.
2.1.3 NQEs and morphosyntactic marking
The work on distributivity and morphosyntactic cues in Serbian (Bosnić 2016) explored the possibility that there may be an impact of verbal agree-ment and nominal inflection on the distributive/collective preference. We conducted two experiments with 70 adults and 49 7-year-old children (25 in Experiment 1 and 24 in Experiment 2) in which we tested whether it is possible that singular verbal agreement would suggest distributive (Winter 2002, Ouwayda 2014) or collective (Wechsler 2009, Haskell & MacDonald 2003) interpretations of an NQE sentence. This was originally hinted by Knežević (2012; 2015) to be a possible explanation for the experimental results of her replication of Musolino’s (2009) study.
The first experiment was done using paucal numerals (numerals dva (two), tri (three) and četiri (four)) that assign a “paucal” case form (syn-cretic to the genitive singular) to the nouns they modify. These paucals can either have a matching verbal agreement (singular) or a mismatching agreement (plural). The second experiment used the the so-called collec-tive mixed-gender numerals (such as dvoje (two), troje (three) and četvoro (four)) which refer to the number of members in sets that contain animate individuals of both sexes (Stanojević 2008). Again, both singular and verbal agreement can be assigned to these mixed-gender numerals. The sentences were numerically quantified in the subject position and the object was a bare singular noun. Adding the type of picture (collective vs. distributive), these combinations resulted in two 2x2 studies, with following conditions (See Figure 4, taken from Bosnić (2016:91):
Collective Distributive Collective Distributive
Tri psa je vuklo/su vukla sanke.
‘Three dogs was pulling/were pulling a sledge.’
Experiment 1
Troje vanzemaljaca nosi/nose merdevine.
‘Three aliens is carrying/are carrying a ladder.’
Experiment 2
Figure 4: Four conditions (two types of pictures x two types of sentences) for Experiments
1 and 2.
Similar to the previous studies (Knežević 2015) adults rejected dis-tributive readings with subject-NQE sentences without any disdis-tributive marking (regardless of the agreement). This proves once again that collective/distributive choice for distributively unmarked sentences is not a matter of preferences, but seems to be a much stronger principle
(almost semantic). This is unlike the English-speaking adults in the studies by Musolino (2009) and Syrett & Musolino (2013), where unmarked sentences were still considered compatible with distributive situations. We assumed here that Serbian adults may be aware of a better way to convey distributive meaning (with po, with svaki or with both) and that distributive readings are simply wrong with distributively unmarked sentences. Children, on the other hand, found both types of pictures highly acceptable, regardless of the agreement and numerals (ranging between 84.67% and 99.3%). Here too, we see that children do not show the same strict response pattern like the adults and are similar to the previous acquisition results (see Table 2). While there was no evidence that morphosyntax plays a role in collective/distribut•ive choices, there was, once again, a confirmation that adults and children have similar patterns of responses like earlier studies in Serbian.
Finally, Serbian allows a construction in which the DistShare marker po can attach to one argument while the remaining argument (in a transitive sentence) is quantified by the DistKey marker svaki. Interestingly, this construction results in individual-distributive and exhaustive readings, that children initially have problems understanding. We present a study investigating Serbian children’s understanding of this construction in the following section.
2.1.4 Svaki and po – individually exhaustive readings
Knežević & Demirdache (2018) investigated comprehension of doubly quantified/marked sentences – with svaki and po using a picture verifica-tion task. Their predicverifica-tions were that the combinaverifica-tion of the two markers would block collective readings (which po alone can do, but not necessarily
svaki (see Knežević (2015: 21) and enforce both exhaustivity (i.e.,
exhaus-tively using the DistKey set) and atomicity, i.e., individual-distributive readings (with 1-to-1 pairing), which are the core properties of universal quantifiers such as svaki, but not attributed to po,6 yielding results similar
to English each.
A total of 31 adults and 98 children participated in this study. Children were divided into three age groups – 22 5-year-olds, 38 7-year-olds and 37 9-year-olds. There were 6 experimental conditions (types of pictures) paired with one type of test sentence containing svaki and po. See Figure 5 (Knežević & Demirdache 2018:129) and example (6) for an overview of the experimental conditions:
(6) Svaka devojka farba po kutiju. every.nom.f.sg girl.nom.sg paint.3.sg distr box.acc.f.sg ‘Each girl is painting a (different) box.’
Adults responded as predicted – they accepted only distributive ex-haustive and atomic scenarios (Fig 5a). The children also had a very high
2
readings with NQEs. The potential influence of morphosyntactic cues ondistributivity acceptance rates has been investigated experimentally with adults and children in Bosnić (2016), and the results are presented in the following section.
2.1.3 NQEs and morphosyntactic marking
The work on distributivity and morphosyntactic cues in Serbian (Bosnić 2016) explored the possibility that there may be an impact of verbal agree-ment and nominal inflection on the distributive/collective preference. We conducted two experiments with 70 adults and 49 7-year-old children (25 in Experiment 1 and 24 in Experiment 2) in which we tested whether it is possible that singular verbal agreement would suggest distributive (Winter 2002, Ouwayda 2014) or collective (Wechsler 2009, Haskell & MacDonald 2003) interpretations of an NQE sentence. This was originally hinted by Knežević (2012; 2015) to be a possible explanation for the experimental results of her replication of Musolino’s (2009) study.
The first experiment was done using paucal numerals (numerals dva (two), tri (three) and četiri (four)) that assign a “paucal” case form (syn-cretic to the genitive singular) to the nouns they modify. These paucals can either have a matching verbal agreement (singular) or a mismatching agreement (plural). The second experiment used the the so-called collec-tive mixed-gender numerals (such as dvoje (two), troje (three) and četvoro (four)) which refer to the number of members in sets that contain animate individuals of both sexes (Stanojević 2008). Again, both singular and verbal agreement can be assigned to these mixed-gender numerals. The sentences were numerically quantified in the subject position and the object was a bare singular noun. Adding the type of picture (collective vs. distributive), these combinations resulted in two 2x2 studies, with following conditions (See Figure 4, taken from Bosnić (2016:91):
Collective Distributive Collective Distributive
Tri psa je vuklo/su vukla sanke.
‘Three dogs was pulling/were pulling a sledge.’
Experiment 1
Troje vanzemaljaca nosi/nose merdevine.
‘Three aliens is carrying/are carrying a ladder.’
Experiment 2
Figure 4: Four conditions (two types of pictures x two types of sentences) for Experiments
1 and 2.
Similar to the previous studies (Knežević 2015) adults rejected dis-tributive readings with subject-NQE sentences without any disdis-tributive marking (regardless of the agreement). This proves once again that collective/distributive choice for distributively unmarked sentences is not a matter of preferences, but seems to be a much stronger principle
(almost semantic). This is unlike the English-speaking adults in the studies by Musolino (2009) and Syrett & Musolino (2013), where unmarked sentences were still considered compatible with distributive situations. We assumed here that Serbian adults may be aware of a better way to
convey distributive meaning (with po, with svaki or with both) and that distributive readings are simply wrong with distributively unmarked sentences. Children, on the other hand, found both types of pictures highly acceptable, regardless of the agreement and numerals (ranging between 84.67% and 99.3%). Here too, we see that children do not show the same strict response pattern like the adults and are similar to the previous acquisition results (see Table 2). While there was no evidence that morphosyntax plays a role in collective/distribut•ive choices, there was, once again, a confirmation that adults and children have similar patterns of responses like earlier studies in Serbian.
Finally, Serbian allows a construction in which the DistShare marker po can attach to one argument while the remaining argument (in a transitive sentence) is quantified by the DistKey marker svaki. Interestingly, this construction results in individual-distributive and exhaustive readings, that children initially have problems understanding. We present a study investigating Serbian children’s understanding of this construction in the following section.
2.1.4 Svaki and po – individually exhaustive readings
Knežević & Demirdache (2018) investigated comprehension of doubly quantified/marked sentences – with svaki and po using a picture verifica-tion task. Their predicverifica-tions were that the combinaverifica-tion of the two markers would block collective readings (which po alone can do, but not necessarily
svaki (see Knežević (2015: 21) and enforce both exhaustivity (i.e.,
exhaus-tively using the DistKey set) and atomicity, i.e., individual-distributive readings (with 1-to-1 pairing), which are the core properties of universal quantifiers such as svaki, but not attributed to po,6 yielding results similar
to English each.
A total of 31 adults and 98 children participated in this study. Children were divided into three age groups – 22 5-year-olds, 38 7-year-olds and 37 9-year-olds. There were 6 experimental conditions (types of pictures) paired with one type of test sentence containing svaki and po. See Figure 5 (Knežević & Demirdache 2018:129) and example (6) for an overview of the experimental conditions:
(6) Svaka devojka farba po kutiju. every.nom.f.sg girl.nom.sg paint.3.sg distr box.acc.f.sg ‘Each girl is painting a (different) box.’
Adults responded as predicted – they accepted only distributive ex-haustive and atomic scenarios (Fig 5a). The children also had a very high
2
rate of acceptance for this condition. However, younger children (5- and7-year-olds) also incorrectly accepted all other conditions (collective, non-exhaustive and non-atomic scenarios) above chance, while 9-year-olds were close to adult-like responses (except accepting non-atomic exhausted scenarios (e.g., Fig 5b) at a higher rate than adults). Only at age 9 were the children able to understand the distributive force of po (which is shown by their rejection of collective situations with po) and their exhaustive requirements with svaki.
Knežević & Demirdache (2018) then conclude that the results seem to suggest that exhaustivity is acquired before atomicity and the truth conditions of po are acquired prior to the truth conditions of svaki. While the first conclusion holds from the results, the second one is harder to infer, since it is difficult to tease apart the semantic contributions of svaki and
po independently, when the test sentence was svaki+po. Their hypothesis
that svaki+po sentences only yield atomic and exhaustive distributive readings holds only for the adults, while the children overall show a delay in the acquisition of both svaki and po. We come back to these results and conclusions specifically in the discussion in section 4.3.
So far, we have seen that we have several studies looking at Serbian children’s understanding of distributivity and distributive marking. To recap, Serbian children do not seem to understand po until the age of 10, but they strongly prefer distributive readings for unmarked sentences. Moreover, it is claimed that in case of the svaki+po combination, Serbian children may understand the truth conditions of po before svaki. However, we still lack data in Serbian on the interpretations of the quantifier svaki in subject position, without po. Such a study would be maximally comparable to many other studies in other languages that looked at DistKey markers (universal quantifiers).
a: atomic-distributive b: non-atomic-distributive c: collective
d: atomic-distributive +
extra subject e: non-atomic-distributive + extra subject f: collective + extra subject
Figure 5: Experimental conditions reflecting exhaustive (a, b, c) and non-exhaustive (d, e, f)
scenarios, together with atomic (a, d) and non-atomic (b, e) scenarios for the test sentence in (6). Collective scenarios (c, f) should be rejected due to the distributive force of po alone.
2.2 Using a new task
We can identify two possible shortcomings with the previous studies – first, all these studies were either picture verification or preference tasks with no additional evidence about the children’s reasoning or explanations of their responses. For that reason, we developed an act-out task7 to uncover
default interpretations children have for markers of distributivity as well as their underlying reasoning in a comprehension task. We believe this task may be more informative to the picture verification task because it does not limit children’s interpretation by restricting answers to yes and no without additional information (unless explicitly prompted) that could uncover children’s reasoning and their acquisition paths to adult-like preferences. On the other hand, by simply asking children to act a situation out, we gain more insights into their response strategies, preferences and some degree of (non-verbal or verbal) reasoning.
Second, for Serbian, as far as we know, there are no studies that tested NQEs, DistShare markers (po) and quantifiers (svaki) with the same children. In addition, by testing children who speak another language (specifically a language with only DistKey markers) with similar materials could offer a good comparison between different languages. Thus, the current experiment will provide the essential information of how the acquisition path differs for children learning a language with and without DistShare markers.
Since the act-out task is rather a novel approach in testing distributivity, we also needed to test it with a language that only has distributive (key) quantifiers, so we can compare with previous research of DistKey markers and to separate language effects from task effects. For this reason, we will also carry out the same task with the same materials with Dutch, a language that has a large body of experimental results related to children’s understanding of quantification and distributivity.
3 Act-out task with Serbian and Dutch children
We used a novel act-out task to test Serbian, a language with both a dis-tributive quantifier (i.e., svaki = every) and a disdis-tributive share marker (i.e.,
po), and Dutch, a language with two universal distributive quantifiers elke
and iedere (=every)8 and no distributive share markers (with findings that
mostly correspond to English data). Recall that the aim of our study is to see how the presence of a DistShare marker, such as po, influences the
7 One other study using an act-out task was carried out by Kiss & Zétényi (2018), who used this method in one of the experiments to show that preschool children are able to perform multiplicative operations prior to being exposed to multiplication through explicit teaching in school, because these multiplica-tion operamultiplica-tions are syntactically encoded in distributively quantified sentences in Hungarian – sentences with a universal quantifier corresponding to every, a distributive key marker corresponding to each and a distributive share marker corresponding to po.
8 Considering the findings that there are no real differences between elke and iedere in Dutch, as reported in several studies (van Koert, 2016; Spenader & Bosnić 2018; de Koster et al. 2017) we excluded iedere from our design and further discussion.
2
rate of acceptance for this condition. However, younger children (5- and7-year-olds) also incorrectly accepted all other conditions (collective, non-exhaustive and non-atomic scenarios) above chance, while 9-year-olds were close to adult-like responses (except accepting non-atomic exhausted scenarios (e.g., Fig 5b) at a higher rate than adults). Only at age 9 were the children able to understand the distributive force of po (which is shown by their rejection of collective situations with po) and their exhaustive requirements with svaki.
Knežević & Demirdache (2018) then conclude that the results seem to suggest that exhaustivity is acquired before atomicity and the truth conditions of po are acquired prior to the truth conditions of svaki. While the first conclusion holds from the results, the second one is harder to infer, since it is difficult to tease apart the semantic contributions of svaki and
po independently, when the test sentence was svaki+po. Their hypothesis
that svaki+po sentences only yield atomic and exhaustive distributive readings holds only for the adults, while the children overall show a delay in the acquisition of both svaki and po. We come back to these results and conclusions specifically in the discussion in section 4.3.
So far, we have seen that we have several studies looking at Serbian children’s understanding of distributivity and distributive marking. To recap, Serbian children do not seem to understand po until the age of 10, but they strongly prefer distributive readings for unmarked sentences. Moreover, it is claimed that in case of the svaki+po combination, Serbian children may understand the truth conditions of po before svaki. However, we still lack data in Serbian on the interpretations of the quantifier svaki in subject position, without po. Such a study would be maximally comparable to many other studies in other languages that looked at DistKey markers (universal quantifiers).
a: atomic-distributive b: non-atomic-distributive c: collective
d: atomic-distributive +
extra subject e: non-atomic-distributive + extra subject f: collective + extra subject
Figure 5: Experimental conditions reflecting exhaustive (a, b, c) and non-exhaustive (d, e, f)
scenarios, together with atomic (a, d) and non-atomic (b, e) scenarios for the test sentence in (6). Collective scenarios (c, f) should be rejected due to the distributive force of po alone.
2.2 Using a new task
We can identify two possible shortcomings with the previous studies – first, all these studies were either picture verification or preference tasks with no additional evidence about the children’s reasoning or explanations of their responses. For that reason, we developed an act-out task7 to uncover
default interpretations children have for markers of distributivity as well as their underlying reasoning in a comprehension task. We believe this task may be more informative to the picture verification task because it does not limit children’s interpretation by restricting answers to yes and no without additional information (unless explicitly prompted) that could uncover children’s reasoning and their acquisition paths to adult-like preferences. On the other hand, by simply asking children to act a situation out, we gain more insights into their response strategies, preferences and some degree of (non-verbal or verbal) reasoning.
Second, for Serbian, as far as we know, there are no studies that tested NQEs, DistShare markers (po) and quantifiers (svaki) with the same children. In addition, by testing children who speak another language (specifically a language with only DistKey markers) with similar materials could offer a good comparison between different languages. Thus, the current experiment will provide the essential information of how the acquisition path differs for children learning a language with and without DistShare markers.
Since the act-out task is rather a novel approach in testing distributivity, we also needed to test it with a language that only has distributive (key) quantifiers, so we can compare with previous research of DistKey markers and to separate language effects from task effects. For this reason, we will also carry out the same task with the same materials with Dutch, a language that has a large body of experimental results related to children’s understanding of quantification and distributivity.
3 Act-out task with Serbian and Dutch children
We used a novel act-out task to test Serbian, a language with both a dis-tributive quantifier (i.e., svaki = every) and a disdis-tributive share marker (i.e.,
po), and Dutch, a language with two universal distributive quantifiers elke
and iedere (=every)8 and no distributive share markers (with findings that
mostly correspond to English data). Recall that the aim of our study is to see how the presence of a DistShare marker, such as po, influences the
7 One other study using an act-out task was carried out by Kiss & Zétényi (2018), who used this method in one of the experiments to show that preschool children are able to perform multiplicative operations prior to being exposed to multiplication through explicit teaching in school, because these multiplica-tion operamultiplica-tions are syntactically encoded in distributively quantified sentences in Hungarian – sentences with a universal quantifier corresponding to every, a distributive key marker corresponding to each and a distributive share marker corresponding to po.
8 Considering the findings that there are no real differences between elke and iedere in Dutch, as reported in several studies (van Koert, 2016; Spenader & Bosnić 2018; de Koster et al. 2017) we excluded iedere from our design and further discussion.
