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.
Non-exhaustive po.
Event plurality marker or
Non-exhaustive po.
Event plurality marker or
5
Abstract
The focus of Chapter 5 is how to semantically analyze non-exhaustive po, that is, po that does not require an exhausted DistKey, as found in one of the populations in the experimental studies from Chapter 4. We explored and compared two possible accounts that do not analyze DistShare as universal quantifiers – the pluractional account proposed by Knežević (2015) for the marker po, for which the only requirement needed is to have at least two events containing a po-marked argument, distributed over spatio-temporal units. The second proposal analyzes DistShare markers as group-forming devices (following McKercher & Kim (1999) and Gil (1990)), which do not necessarily require a plurality of events for allowing felicitous scenarios with DistShare markers. Thus, the two accounts differ in the fact that the latter allows scenarios that involve a single event, which the event plurality account blocks as a possible scenario. We contribute to this argumentation by discussing different criteria for individuating and counting events in order to apply them to the counterexamples to the group-forming claims and argue against single-event scenarios. Furthermore, using our own borderline example involving minimal spatial separation, we demonstrate, in a step-wise manner, that single-event scenarios should be considered as multiple-event scenarios. We argue in favor of the pluractional analysis of non-exhaustive po, since our claim states that if there is at least one dimension over which the events are not overlapping, this counts as a minimal differentiation condition that results in separate (sub)events.
1 Introduction
The previous chapters of the thesis dealt with both children and adult interpretations of distributive markers in Serbian, Korean and Dutch using experimental methods and statistical models. While Chapter 2 established the acquisition path of distributive key (DistKey) and distributive share (DistShare) markers in Serbian and Dutch, Chapters 3 and 4 tested whether DistShare markers are universal quantifiers (UQ) or pluractional markers.
In Chapter 4, in particular, we identified two distinct populations of speakers when testing transitive sentences: one population treats po as a universal distributive quantifier, and the other one does not. The division was based on an exhaustivity requirement on the DistKey, which is a mandatory requirement for universal quantifiers. Thus, the population that requires po to be exhaustive on its DistKey treats po as a universal quantifier, an analysis of DistShare markers that has been extensively argued for by Choe (1987), Zimmermann (2002), Champollion (2012, 2016b) and Przepiórkowski (2008, 2015), among others, and crucially also in Chapter 3. In this chapter, however, we are solely interested in our results for the second population, for which po does not impose an exhaustivity requirement on a DistKey. We explore here the question of how to analyze this other po, which we henceforth refer to as non-exhaustive po.
The novel criterion we used to disambiguate exhaustivity requirements from maximality effects of definite plural expressions, and confirm we are indeed dealing with two populations, was homogeneity. In essence, plural definites are said to give rise to homogeneity effects, i.e., there is a “truth-value gap” in which a statement (where maximality does not hold) is judged as not completely true nor completely false, resulting in trivalence in its truth conditions (Križ & Chemla 2015). Universal quantifiers, on the other hand, do not show homogeneity effects (they strictly require exhaustivity) and the difference between definite plurals and universal quantifiers comes out clearly in negative statements. Specifically, universal quantifiers in negative contexts should be judged as true when paired with the non-exhausted scenarios, and, conversely, statements with definite plural expressions should either be judged as false or as a “gap”, when paired with non-exhausted scenarios (see Chapter 4, section 5, for a detailed overview of homogeneity issues). Thus, sentences with po could either pattern like universal quantifiers or definite plurals and consequently show homogeneity effects.
First, we have to understand the scope of possible interpretations of non-exhaustive po and ask: Is non-exhaustive po really a pluractional marker? In this chapter, we consider two alternative approaches that share the assumption that the semantics of these markers does not involve UQ, but nonetheless differ with respect to the issue whether there is an event plurality requirement – the pluractional account that has a plurality of event requirement and the group-forming account that does not.
Having two populations with distinct interpretations usually signals semantic change, which is what we suggest for po in Chapter 4. We also
5
Abstract
The focus of Chapter 5 is how to semantically analyze non-exhaustive po, that is, po that does not require an exhausted DistKey, as found in one of the populations in the experimental studies from Chapter 4. We explored and compared two possible accounts that do not analyze DistShare as universal quantifiers – the pluractional account proposed by Knežević (2015) for the marker po, for which the only requirement needed is to have at least two events containing a po-marked argument, distributed over spatio-temporal units. The second proposal analyzes DistShare markers as group-forming devices (following McKercher & Kim (1999) and Gil (1990)), which do not necessarily require a plurality of events for allowing felicitous scenarios with DistShare markers. Thus, the two accounts differ in the fact that the latter allows scenarios that involve a single event, which the event plurality account blocks as a possible scenario. We contribute to this argumentation by discussing different criteria for individuating and counting events in order to apply them to the counterexamples to the group-forming claims and argue against single-event scenarios. Furthermore, using our own borderline example involving minimal spatial separation, we demonstrate, in a step-wise manner, that single-event scenarios should be considered as multiple-event scenarios. We argue in favor of the pluractional analysis of non-exhaustive po, since our claim states that if there is at least one dimension over which the events are not overlapping, this counts as a minimal differentiation condition that results in separate (sub)events.
1 Introduction
The previous chapters of the thesis dealt with both children and adult interpretations of distributive markers in Serbian, Korean and Dutch using experimental methods and statistical models. While Chapter 2 established the acquisition path of distributive key (DistKey) and distributive share (DistShare) markers in Serbian and Dutch, Chapters 3 and 4 tested whether DistShare markers are universal quantifiers (UQ) or pluractional markers. In Chapter 4, in particular, we identified two distinct populations of speakers when testing transitive sentences: one population treats po as a universal distributive quantifier, and the other one does not. The division was based on an exhaustivity requirement on the DistKey, which is a mandatory requirement for universal quantifiers. Thus, the population that requires po to be exhaustive on its DistKey treats po as a universal quantifier, an analysis of DistShare markers that has been extensively argued for by Choe (1987), Zimmermann (2002), Champollion (2012, 2016b) and Przepiórkowski (2008, 2015), among others, and crucially also
in Chapter 3. In this chapter, however, we are solely interested in our results for the second population, for which po does not impose an exhaustivity requirement on a DistKey. We explore here the question of how to analyze this other po, which we henceforth refer to as non-exhaustive po.
The novel criterion we used to disambiguate exhaustivity requirements from maximality effects of definite plural expressions, and confirm we are indeed dealing with two populations, was homogeneity. In essence, plural definites are said to give rise to homogeneity effects, i.e., there is a “truth-value gap” in which a statement (where maximality does not hold) is judged as not completely true nor completely false, resulting in trivalence in its truth conditions (Križ & Chemla 2015). Universal quantifiers, on the other hand, do not show homogeneity effects (they strictly require exhaustivity) and the difference between definite plurals and universal quantifiers comes out clearly in negative statements. Specifically, universal quantifiers in negative contexts should be judged as true when paired with the non-exhausted scenarios, and, conversely, statements with definite plural expressions should either be judged as false or as a “gap”, when paired with non-exhausted scenarios (see Chapter 4, section 5, for a detailed overview of homogeneity issues). Thus, sentences with po could either pattern like universal quantifiers or definite plurals and consequently show homogeneity effects.
First, we have to understand the scope of possible interpretations of non-exhaustive po and ask: Is non-exhaustive po really a pluractional marker? In this chapter, we consider two alternative approaches that share the assumption that the semantics of these markers does not involve UQ, but nonetheless differ with respect to the issue whether there is an event plurality requirement – the pluractional account that has a plurality of event requirement and the group-forming account that does not.
Having two populations with distinct interpretations usually signals semantic change, which is what we suggest for po in Chapter 4. We also
5
speculate that po may have lost its universal quantificational force and behaves like a marker of event plurality. In other words, it is undergoing a semantic weakening. We back up this speculation with the results from Chapter 3, in which we only found evidence in favor or universal quan-tification analysis. However, at this stage, we cannot rule out an analysis based on the strengthening of po – i.e., that po starts off as a weaker, non-ex-haustive marker, and becomes a universal quantifier. This uncertainty largely comes from the lack of experimental data on homogeneity effects with intransitive po-sentences and event-distributive (group) readings, which could provide useful data to the discussion of the directionality of semantic change.
This chapter is organized as follows: Section 2 elaborates on the two alternative approaches that may explain non-exhaustive po. The first is the event plurality account developed specifically for Serbian po by Knežević (2015) in which yielding multiple events is the core property of DistShare markers. On this proposal, distributive readings are obligatory for DistShare markers (section 2.1).1 The second approach is put forth
by McKercher & Kim (1999) for Korean -ssik, and their claim is that DistShare markers are “group-forming devices” that involve a plurality of entities/individuals but not necessarily multiple events. As such, these group-forming devices do not have distributivity as part of their meaning, i.e., distributive readings are then optional (section 2.2). By compar-ing the two approaches, we revisit the principles for individuatcompar-ing and counting events and provide evidence that we are, in fact, dealing with event plurality. We argue that event plurality must necessarily follow if there is at least one dimension over which distribution takes place (individuals and/or times and/or spaces) (section 3). To resolve the issue at stake here, i.e., whether po distributes over events or not, we introduce our own borderline example involving minimal spatial separation of events (with crucially no separation along either the temporal or the theta role dimensions). We use this borderline case of minimal separation as a diagnostic to differentiate between sub-events and to support the event plurality approach for non-exhaustive po (section 4). Finally, we conclude the chapter by highlighting the importance of considering all three dimensions (individuals, time and space) over which distribution takes place when judging sentences with po (section 5).
2 Event plurality vs. Group-forming device
There are two alternative approaches to the UQ analysis that could account for non-exhaustive po. The core difference between the two is twofold: the questions are (i) whether there is an event plurality requirement or not, and (ii) whether distributivity is a byproduct of event plurality or is independent of it.
1 Even though we proposed an event plurality account in Chapter 4, we will work out what this would mean in more detail in this chapter.
Specifically, on the event plurality account, distributive effects are “claimed to be a byproduct of the pairing of sub-events along three possible dimensions: temporal intervals, spatial locations or partici-pants” (Donazzan & Müller 2015: 124). On the group-forming account, however, the meaning of DistShare markers is “independent of whatever semantic mechanism gives rise to distributive readings”, and as such can be compatible with single event readings (McKercher & Kim 1999: 247). We put forth empirical evidence (sentences under different contexts) in order to determine if po can indeed be used in single event scenarios, as predicted by the group-forming proposal. We show that the key examples used to support this proposal over the event plurality account can be construed as multiple sub-events, minimally distributed over one of the three dimensions. Given these examples, we conclude that the arguments for the group-forming proposal are not sustainable, and we endorse the event plurality account.
2.1 Plurality of events
To reiterate from Chapter 3, a pluractional is a term coined by Newman (1980) to describe the variety of lexical items cross-linguistically that languages use to mark the multiplicity of (sufficiently similar)2 events
on a verb or a verb phrase. The multiple events need to involve (and be partitioned into) multiple occasions (times), spaces and/or participants (Newman 1980, Lasersohn 1995, Cabredo-Hofherr & Tovena 2015). Many plural meanings are thus associated with pluractionals: for instance, Cusic (1981:74), who worked extensively on verbal plurality, listed meanings like repetitiveness, repeated occasions or events, habitual agency, dis-tributed quality, inchoativity, distribution, cumulative result, intensity, augmentation, diminuation, plurality of sites of action, duration, persistent consequences, celerativity or continuity, all as possible types of plural verbal meanings that can be marked/signaled by pluractional markers. In languages such as English, markers of event plurality are verbs such as
rewrite or nibble. Other subcategories of markers of event plurality also
include adverbials (always, often) and adnominal markers (different, each) (Cabredo-Hofherr & Laca 2012; Cabredo-Hofherr 2010).
Distributivity can be expressed via different mechanisms – via scope-tak-ing elements, cumulative denotations (scopeless readscope-tak-ings related to collec-tive, but necessarily resulting in plurality of events (see Scha 1981; Winter 2000; Champollion 2017),3 or that distributivity could simply be inherent
to some event plurality markers (pluractionals) (see Henderson 2012; Cable 2014; Donazzan and Müller 2015). More specifically, according to
2 The logical question arises – what does it exactly mean to be sufficiently similar? We take on this issue in sections 3 and 4 in much more detail.
3 Cumulativity is a phenomenon widely discussed in the literature, with different approaches as to how to analyze it (c.f. Krifka 1992, 1999; Roberts 1987; Beck & Sauerland 2000; Kratzer 2007; Champollion 2016a). Cumulativity has not been investigated experimentally in this thesis so we will not go in a more detailed discussion of the phenomenon. However, it is worth noting that the analysis of cumulative partitioning is also one of the differences between Knežević’s (2015) and Balusu’s (2006) accounts in terms of how the unmarked argument is distributed over the partitioned events.
5
speculate that po may have lost its universal quantificational force and behaves like a marker of event plurality. In other words, it is undergoing a semantic weakening. We back up this speculation with the results from Chapter 3, in which we only found evidence in favor or universal quan-tification analysis. However, at this stage, we cannot rule out an analysis based on the strengthening of po – i.e., that po starts off as a weaker, non-ex-haustive marker, and becomes a universal quantifier. This uncertainty largely comes from the lack of experimental data on homogeneity effects with intransitive po-sentences and event-distributive (group) readings, which could provide useful data to the discussion of the directionality of semantic change.
This chapter is organized as follows: Section 2 elaborates on the two alternative approaches that may explain non-exhaustive po. The first is the event plurality account developed specifically for Serbian po by Knežević (2015) in which yielding multiple events is the core property of DistShare markers. On this proposal, distributive readings are obligatory for DistShare markers (section 2.1).1 The second approach is put forth
by McKercher & Kim (1999) for Korean -ssik, and their claim is that DistShare markers are “group-forming devices” that involve a plurality of entities/individuals but not necessarily multiple events. As such, these group-forming devices do not have distributivity as part of their meaning, i.e., distributive readings are then optional (section 2.2). By compar-ing the two approaches, we revisit the principles for individuatcompar-ing and counting events and provide evidence that we are, in fact, dealing with event plurality. We argue that event plurality must necessarily follow if there is at least one dimension over which distribution takes place (individuals and/or times and/or spaces) (section 3). To resolve the issue at stake here, i.e., whether po distributes over events or not, we introduce our own borderline example involving minimal spatial separation of events (with crucially no separation along either the temporal or the theta role dimensions). We use this borderline case of minimal separation as a diagnostic to differentiate between sub-events and to support the event plurality approach for non-exhaustive po (section 4). Finally, we conclude the chapter by highlighting the importance of considering all three dimensions (individuals, time and space) over which distribution takes place when judging sentences with po (section 5).
2 Event plurality vs. Group-forming device
There are two alternative approaches to the UQ analysis that could account for non-exhaustive po. The core difference between the two is twofold: the questions are (i) whether there is an event plurality requirement or not, and (ii) whether distributivity is a byproduct of event plurality or is independent of it.
1 Even though we proposed an event plurality account in Chapter 4, we will work out what this would mean in more detail in this chapter.
Specifically, on the event plurality account, distributive effects are “claimed to be a byproduct of the pairing of sub-events along three possible dimensions: temporal intervals, spatial locations or partici-pants” (Donazzan & Müller 2015: 124). On the group-forming account, however, the meaning of DistShare markers is “independent of whatever semantic mechanism gives rise to distributive readings”, and as such can be compatible with single event readings (McKercher & Kim 1999: 247). We put forth empirical evidence (sentences under different contexts) in order to determine if po can indeed be used in single event scenarios, as predicted by the group-forming proposal. We show that the key examples used to support this proposal over the event plurality account can be construed as multiple sub-events, minimally distributed over one of the three dimensions. Given these examples, we conclude that the arguments for the group-forming proposal are not sustainable, and we endorse the event plurality account.
2.1 Plurality of events
To reiterate from Chapter 3, a pluractional is a term coined by Newman (1980) to describe the variety of lexical items cross-linguistically that languages use to mark the multiplicity of (sufficiently similar)2 events
on a verb or a verb phrase. The multiple events need to involve (and be partitioned into) multiple occasions (times), spaces and/or participants (Newman 1980, Lasersohn 1995, Cabredo-Hofherr & Tovena 2015). Many plural meanings are thus associated with pluractionals: for instance, Cusic (1981:74), who worked extensively on verbal plurality, listed meanings like repetitiveness, repeated occasions or events, habitual agency, dis-tributed quality, inchoativity, distribution, cumulative result, intensity, augmentation, diminuation, plurality of sites of action, duration, persistent consequences, celerativity or continuity, all as possible types of plural verbal meanings that can be marked/signaled by pluractional markers. In languages such as English, markers of event plurality are verbs such as
rewrite or nibble. Other subcategories of markers of event plurality also
include adverbials (always, often) and adnominal markers (different, each) (Cabredo-Hofherr & Laca 2012; Cabredo-Hofherr 2010).
Distributivity can be expressed via different mechanisms – via scope-tak-ing elements, cumulative denotations (scopeless readscope-tak-ings related to collec-tive, but necessarily resulting in plurality of events (see Scha 1981; Winter 2000; Champollion 2017),3 or that distributivity could simply be inherent
to some event plurality markers (pluractionals) (see Henderson 2012; Cable 2014; Donazzan and Müller 2015). More specifically, according to
2 The logical question arises – what does it exactly mean to be sufficiently similar? We take on this issue in sections 3 and 4 in much more detail.
3 Cumulativity is a phenomenon widely discussed in the literature, with different approaches as to how to analyze it (c.f. Krifka 1992, 1999; Roberts 1987; Beck & Sauerland 2000; Kratzer 2007; Champollion 2016a). Cumulativity has not been investigated experimentally in this thesis so we will not go in a more detailed discussion of the phenomenon. However, it is worth noting that the analysis of cumulative partitioning is also one of the differences between Knežević’s (2015) and Balusu’s (2006) accounts in terms of how the unmarked argument is distributed over the partitioned events.
5
Donazzan & Müller (2015: 114), distributive numerals (numerals marked as DistShare) “enforce a distributive relation between the events in the ex-tension of the distributive share and some external key of distribution pro-vided by participants, temporal intervals and spatial locations.” However, they showed that reduplicated numerals (at least in Mandarin Chinese and Karitiana) (i) do not depend on a distributive operator and the existence of an overt licensor (as is the case of dependent indefinites) and that (ii) the potential DistKey need not exhaust all the entities in the domain, which is in direct contradiction to the hypothesis that DistShare markers are universal quantifiers. In fact, following Cable (2014), Donazzan & Müller (2015) argue that reduplicated numerals are pluractional operators and that distributivity is necessarily a byproduct of the pluractionalization of the predicate along three possible dimensions – time, space and participants/ individuals.
Authors such as Knežević (2015) (who also followed Cable 2014) adopted the pluractional analysis of the DistShare marker po. The authors claim that these markers are that they are not universal quantifiers, but instead that they simply entail a plural (sum) of events. We tested the first claim in Chapters 3 and 4, namely, we tested whether these DistShare markers have exhaustivity requirements that would classify them as universal distributive quantifiers, as others have argued (e.g., Choe 1987, Gil 1995, Zimmer-mann 2002, Balusu 2006, a.o.). After identifying the two populations and that interpretation of po yields two different possibilities: exhaustive and non-exhaustive, we should now test the second claim – the plurality of event requirement. Apart from Knežević’s analysis of po, a pluractional analysis of event-distributive markers has also been pursued in recent studies in languages like Karitiana (Muller & Negrão 2012), St’at’imcets (Matthewson 2000), Kaqchikel (Henderson 2014), Tlingit (Cable 2014) and Seri (Cabredo-Hofherr et al. 2019). Although Cable’s proposal has directly influenced Knežević’s analysis of po (both accounts were outlined in Chapter 3, section 2.2), we take the next section (3.1.1) to highlight the similarities and a crucial difference between Cable and Knežević – the covariation requirement.
2.1.1 Cable vs. Knežević
We have already extensively presented Cable’s (2014) and Knežević’s (2015) in Chapter 3 in which we showed how their semantics derive distributive meanings. In this section, however, we will point out some differences between these accounts.
The semantics and truth conditions of -gaa in Tlingit (Cable 2014: 586) and po in Serbian (Knežević & Demirdache 2018) are repeated in (1) and (2) respectively:
(1) a. Ax shaa yátx’i dáxgaa keitl has aawashúch.
my female children two.distr dog they.bathed
b. [[ gaa ]] = [ λnn : [ λQ<et> : [ λP<e, εt> : [ λeε: ∃x. Q(x) & P(x)(e) &
<e, x> = σ<e’, y>. participant(e’,y) & |y| = n & e’ < e & y < x ] … ]
c. ∃e.∃x.*dog(x) & *bathed(e) & *Agent(e) = σy.*my.daughter(y) & *Theme(e) = x & <e , x> = σ<e’, z> . z < x & |z| = 2 & e’ < e & participant(e’,z)
d. There is a plural event e of bathing, whose agents are the speaker’s daugh-ters and whose theme is a plurality of dogs x and the pair consisting of e and x is the sum of those pairs <e’, z> such that z is a pair of dogs, e’ is a (proper) part of e, and z participates in e’.
(2) a. Moje ćerke kupaju po dva psa.
my daughters bathe distr two dogs.acc
b. [[po]] = λn. λQ<e,t>. λP<e, εt >. λe. e ∈ *e’nQ & e ∉ e’nQ & ∃x Q(x) & P(x)(e)
c. ∃e. ∃y. ∃x. e ∈ *e’2dogs & e ∉ e’2dogs & *my.daughter(y) & *dog(x) &
*bathe(e)
& *Agent(e)(y) & *Theme(e)(x)
d. There is an event e constructed out of (sub)events e’ each involving two dogs and e is an event of the speaker’s daughters cumulatively bathing dogs.
This brings us to the crucial difference between Knežević and Cable – Knežević does not have a covariation requirement in her semantics, while Cable does. Let’s see how this requirement is built into Cable’s system: In Tlingit (see (1)), the adnominal distributive numeral -gaa must combine with the numNP and then it takes a relation P between entities and events, and returns a predicate of events. One of the most important properties of -gaa is that y must be a proper part of x (represented as y < x). Recall that x is equivalent to the plural expression that introduces the domain out of which y is taken (i.e., x = plurality of dogs, y = pairs of dogs). Being a proper part of x (dogs) then means that y (pairs of dogs) must covary, that is, there must be different pairs of dogs involved in the bathing subevents. Additionally, the pair <e,x> is the sum of the pairs <e’,y> such that the subevent e’ must be a proper part of event e (represented as e’ < e), meaning that there is no trivial partitioning of events in which the whole event is partitioned in a single subevent (contra Balusu (2006), for instance).
Although it is natural to assume covariation of the members/units of the DistShare argument (also observable crosslinguistically, see e.g., Far-kas 1997, 2015; Henderson 2012), Knežević presents potential empirical arguments both for and against the covariation requirement for po. On the one hand, Knežević’s informants indeed dispreferred the situation in which the same pairs of dogs are being bathed over a certain period of time (say every week the daughters are bathing the same two dogs). In other words, dogs should covary per subevent. However, it is still not clear how to deal with intermediate cases, i.e., can some dogs be the same and to what degree is this acceptable or not? In addition, po combines with indefinite expressions and as such should be considered non-specific, which further implies that there should be no re-using of the members per subevent (see also Zimmermann 2002). Again, we do not know to what extent this argument holds. On the other hand, however, there are contexts that may not require covariation. For example, say there are several football games
5
Donazzan & Müller (2015: 114), distributive numerals (numerals marked as DistShare) “enforce a distributive relation between the events in the ex-tension of the distributive share and some external key of distribution pro-vided by participants, temporal intervals and spatial locations.” However, they showed that reduplicated numerals (at least in Mandarin Chinese and Karitiana) (i) do not depend on a distributive operator and the existence of an overt licensor (as is the case of dependent indefinites) and that (ii) the potential DistKey need not exhaust all the entities in the domain, which is in direct contradiction to the hypothesis that DistShare markers are universal quantifiers. In fact, following Cable (2014), Donazzan & Müller (2015) argue that reduplicated numerals are pluractional operators and that distributivity is necessarily a byproduct of the pluractionalization of the predicate along three possible dimensions – time, space and participants/ individuals.
Authors such as Knežević (2015) (who also followed Cable 2014) adopted the pluractional analysis of the DistShare marker po. The authors claim that these markers are that they are not universal quantifiers, but instead that they simply entail a plural (sum) of events. We tested the first claim in Chapters 3 and 4, namely, we tested whether these DistShare markers have exhaustivity requirements that would classify them as universal distributive quantifiers, as others have argued (e.g., Choe 1987, Gil 1995, Zimmer-mann 2002, Balusu 2006, a.o.). After identifying the two populations and that interpretation of po yields two different possibilities: exhaustive and non-exhaustive, we should now test the second claim – the plurality of event requirement. Apart from Knežević’s analysis of po, a pluractional analysis of event-distributive markers has also been pursued in recent studies in languages like Karitiana (Muller & Negrão 2012), St’at’imcets (Matthewson 2000), Kaqchikel (Henderson 2014), Tlingit (Cable 2014) and Seri (Cabredo-Hofherr et al. 2019). Although Cable’s proposal has directly influenced Knežević’s analysis of po (both accounts were outlined in Chapter 3, section 2.2), we take the next section (3.1.1) to highlight the similarities and a crucial difference between Cable and Knežević – the covariation requirement.
2.1.1 Cable vs. Knežević
We have already extensively presented Cable’s (2014) and Knežević’s (2015) in Chapter 3 in which we showed how their semantics derive distributive meanings. In this section, however, we will point out some differences between these accounts.
The semantics and truth conditions of -gaa in Tlingit (Cable 2014: 586) and po in Serbian (Knežević & Demirdache 2018) are repeated in (1) and (2) respectively:
(1) a. Ax shaa yátx’i dáxgaa keitl has aawashúch.
my female children two.distr dog they.bathed
b. [[ gaa ]] = [ λnn : [ λQ<et> : [ λP<e, εt> : [ λeε: ∃x. Q(x) & P(x)(e) &
<e, x> = σ<e’, y>. participant(e’,y) & |y| = n & e’ < e & y < x ] … ]
c. ∃e.∃x.*dog(x) & *bathed(e) & *Agent(e) = σy.*my.daughter(y) & *Theme(e) = x & <e , x> = σ<e’, z> . z < x & |z| = 2 & e’ < e & participant(e’,z)
d. There is a plural event e of bathing, whose agents are the speaker’s daugh-ters and whose theme is a plurality of dogs x and the pair consisting of e and x is the sum of those pairs <e’, z> such that z is a pair of dogs, e’ is a (proper) part of e, and z participates in e’.
(2) a. Moje ćerke kupaju po dva psa.
my daughters bathe distr two dogs.acc
b. [[po]] = λn. λQ<e,t>. λP<e, εt >. λe. e ∈ *e’nQ & e ∉ e’nQ & ∃x Q(x) & P(x)(e)
c. ∃e. ∃y. ∃x. e ∈ *e’2dogs & e ∉ e’2dogs & *my.daughter(y) & *dog(x) &
*bathe(e)
& *Agent(e)(y) & *Theme(e)(x)
d. There is an event e constructed out of (sub)events e’ each involving two dogs and e is an event of the speaker’s daughters cumulatively bathing dogs.
This brings us to the crucial difference between Knežević and Cable – Knežević does not have a covariation requirement in her semantics, while Cable does. Let’s see how this requirement is built into Cable’s system: In Tlingit (see (1)), the adnominal distributive numeral -gaa must combine with the numNP and then it takes a relation P between entities and events, and returns a predicate of events. One of the most important properties of -gaa is that y must be a proper part of x (represented as y < x). Recall that x is equivalent to the plural expression that introduces the domain out of which y is taken (i.e., x = plurality of dogs, y = pairs of dogs). Being a proper part of x (dogs) then means that y (pairs of dogs) must covary, that is, there must be different pairs of dogs involved in the bathing subevents. Additionally, the pair <e,x> is the sum of the pairs <e’,y> such that the subevent e’ must be a proper part of event e (represented as e’ < e), meaning that there is no trivial partitioning of events in which the whole event is partitioned in a single subevent (contra Balusu (2006), for instance).
Although it is natural to assume covariation of the members/units of the DistShare argument (also observable crosslinguistically, see e.g., Far-kas 1997, 2015; Henderson 2012), Knežević presents potential empirical arguments both for and against the covariation requirement for po. On the one hand, Knežević’s informants indeed dispreferred the situation in which the same pairs of dogs are being bathed over a certain period of time (say every week the daughters are bathing the same two dogs). In other words, dogs should covary per subevent. However, it is still not clear how to deal with intermediate cases, i.e., can some dogs be the same and to what degree is this acceptable or not? In addition, po combines with indefinite expressions and as such should be considered non-specific, which further implies that there should be no re-using of the members per subevent (see also Zimmermann 2002). Again, we do not know to what extent this argument holds. On the other hand, however, there are contexts that may not require covariation. For example, say there are several football games
5
during one week in which one referee raised the yellow card twice per game. One could say – During the games, this referee raised po two yellow
cards, even though it could be that it was exactly the same card that was
raised, but it was aimed at different people. These examples are thus the reason the covariation requirement has been left out for po (for now). At this point, however, it is impossible to say whether the covariation requirement is necessary for the semantics of (all) DistShare markers. In addition, it is not entirely clear if and how it can affect the individuation of events over individuals which we discuss in section 3 below. Although we come back to this issue in Chapter 6, it is important to highlight that the topic of covariation (and explicit difference between Knežević and Cable) has not been experimentally tested and we encourage future research in this direction.
2.2 Group-forming device
An interesting alternative analysis (though not extensively pursued in the literature) is that of McKercher & Kim (1999) who argue that DistShare markers are not pluractionals (and not even distributive markers), but group-forming devices.4 More precisely, they argue that forming multiple
groups, although involving a plurality of individuals/groups, does not require distributivity and it does not necessarily result in plurality of events. Since we have established semantic similarities between -ssik and po, and also shown that experimentally they behave the same (see Chapter 3), we will give a critical analysis of McKercher & Kim’s semantics and try to see if it would be compatible with po and perhaps other DistShare markers that behave similarly.
2.2.1 Korean -ssik – type-lifting to group-forming
McKercher & Kim (1999) claim that -ssik simply denotes multiple groups, so that -ssik is actually a “group former” that “combines with the meaning of a quantifier phrase and returns a higher type denotation which means GROUPS of [numeral] n” (ibid:243). We break down their reasoning and their semantics below.
Specifically, they argue that -ssik overtly marks the type-lifting operation of a determiner from the first type (a relation between sets) to the second type (a relation between a set and a set of sets) – i.e., “a set of elements is lifted to a set of sets of elements. In other words, groups are formed. Implicit in this idea of group formation is the requirement that the size of the groups be specified” (McKercher & Kim 1999: 243). This has been discussed in Verkuyl (1994) and van der Does (1993), who also dealt with distributive and collective readings. In particular, Verkuyl proposes that there should not be a polarized distinction between collective and
4 Another author that adopts the group-forming analysis is Faller (2001) for the Quechua DistShare marker -nka. However, her proposal only has the group-forming device applied to a subset of exam-ples with -nka that involves stative verbs. For the active verbs, she analyses Quechua as a UQ marker that enters a Key-Share relationship. This approach is less desirable because it relies on two distinct proposals being implemented for the same marker. Nevertheless, it would still be very interesting to experimentally test these two conditions of the -nka marker in (Peruvian) Quechua.
distributive readings, but rather a unified notion of a scale of distribu-tivity with collective readings on the one end and atomic (distributive) readings on the other. Partially distributive meanings would then fall in between on the scale. To give a semantic account of this unified version of distributivity, Verkuyl adopted van der Does’ proposal to type-lift the denotation of the determiner phrase. The type-lifted determiner phrase would be able to deal with all these readings, putting the DP as a dominant source of distributivity/collectivity over the VP. McKercher & Kim then follow Verkuyl and van der Does in making use of this type-lifting operation.
Type-lifting for McKercher & Kim means that -ssik (or po) combines with a cardinal (or a weak) quantifier and returns a higher type of a numP which then combines with a noun. The second function of this modified numerically quantified phrase is to multiply the group it created. Consider the proposed semantics of -ssik in (3) (ibid:244):
(3) -ssik => λDλXλR. Condition 1 & Condition 2
• D is of type ≪e,t>,≪e,t>,t≫ (a quantifier relation), the denotation of the cardinal quantifier that combines with -ssik
• X is of type <e,t> (a set of elements), the denotation of the noun
• R is of type ≪e,t>,t> (a set of sets), the type-lifted denotation of the verb phrase
Condition 1: |R∩{Z:Z∩X≠0}|≥2 (there must be multiple groups)
Condition 2: ∀Z ∈R(Z∩X≠0 →D(X)(Z)) (the cardinal determines size of participants in group)
Importantly, note that Condition 1 looks like a plurality condition – it creates multiple groups, but plural groups do not necessarily entail plurality of events. McKercher & Kim (1999) explicitly argue that plurality of events and plurality of groups could not be the same, since they claim -ssik is compatible with single events. Thus, plurality of groups and plurality of events yield different predictions.
In order to see how these conditions apply to Serbian to get the intended meaning, we will use a simple po-sentence in (4) (a slight variation of the test sentence from Chapter 3):
(4) Plešu po tri majmuna.
dance distr three monkeys
‘distr three monkeys are dancing.’
The cardinal quantifier D in (4) is number three, X is the denotation of
monkeys and R is the denotation of the verb phrase are dancing.
Condi-tion 1, the essential function for the given semantics of -ssik/po, states that the number of sets that contain the members of the noun denotation (monkeys) and that are in the denotation of the verb phrase (are dancing) must be equal or greater than two, i.e., there should be multiple groups of monkeys dancing. Further, Condition 2 gives the number of members
5
during one week in which one referee raised the yellow card twice per game. One could say – During the games, this referee raised po two yellow
cards, even though it could be that it was exactly the same card that was
raised, but it was aimed at different people. These examples are thus the reason the covariation requirement has been left out for po (for now). At this point, however, it is impossible to say whether the covariation requirement is necessary for the semantics of (all) DistShare markers. In addition, it is not entirely clear if and how it can affect the individuation of events over individuals which we discuss in section 3 below. Although we come back to this issue in Chapter 6, it is important to highlight that the topic of covariation (and explicit difference between Knežević and Cable) has not been experimentally tested and we encourage future research in this direction.
2.2 Group-forming device
An interesting alternative analysis (though not extensively pursued in the literature) is that of McKercher & Kim (1999) who argue that DistShare markers are not pluractionals (and not even distributive markers), but group-forming devices.4 More precisely, they argue that forming multiple
groups, although involving a plurality of individuals/groups, does not require distributivity and it does not necessarily result in plurality of events. Since we have established semantic similarities between -ssik and po, and also shown that experimentally they behave the same (see Chapter 3), we will give a critical analysis of McKercher & Kim’s semantics and try to see if it would be compatible with po and perhaps other DistShare markers that behave similarly.
2.2.1 Korean -ssik – type-lifting to group-forming
McKercher & Kim (1999) claim that -ssik simply denotes multiple groups, so that -ssik is actually a “group former” that “combines with the meaning of a quantifier phrase and returns a higher type denotation which means GROUPS of [numeral] n” (ibid:243). We break down their reasoning and their semantics below.
Specifically, they argue that -ssik overtly marks the type-lifting operation of a determiner from the first type (a relation between sets) to the second type (a relation between a set and a set of sets) – i.e., “a set of elements is lifted to a set of sets of elements. In other words, groups are formed. Implicit in this idea of group formation is the requirement that the size of the groups be specified” (McKercher & Kim 1999: 243). This has been discussed in Verkuyl (1994) and van der Does (1993), who also dealt with distributive and collective readings. In particular, Verkuyl proposes that there should not be a polarized distinction between collective and
4 Another author that adopts the group-forming analysis is Faller (2001) for the Quechua DistShare marker -nka. However, her proposal only has the group-forming device applied to a subset of exam-ples with -nka that involves stative verbs. For the active verbs, she analyses Quechua as a UQ marker that enters a Key-Share relationship. This approach is less desirable because it relies on two distinct proposals being implemented for the same marker. Nevertheless, it would still be very interesting to experimentally test these two conditions of the -nka marker in (Peruvian) Quechua.
distributive readings, but rather a unified notion of a scale of distribu-tivity with collective readings on the one end and atomic (distributive) readings on the other. Partially distributive meanings would then fall in between on the scale. To give a semantic account of this unified version of distributivity, Verkuyl adopted van der Does’ proposal to type-lift the denotation of the determiner phrase. The type-lifted determiner phrase would be able to deal with all these readings, putting the DP as a dominant source of distributivity/collectivity over the VP. McKercher & Kim then follow Verkuyl and van der Does in making use of this type-lifting operation.
Type-lifting for McKercher & Kim means that -ssik (or po) combines with a cardinal (or a weak) quantifier and returns a higher type of a numP which then combines with a noun. The second function of this modified numerically quantified phrase is to multiply the group it created. Consider the proposed semantics of -ssik in (3) (ibid:244):
(3) -ssik => λDλXλR. Condition 1 & Condition 2
• D is of type ≪e,t>,≪e,t>,t≫ (a quantifier relation), the denotation of the cardinal quantifier that combines with -ssik
• X is of type <e,t> (a set of elements), the denotation of the noun
• R is of type ≪e,t>,t> (a set of sets), the type-lifted denotation of the verb phrase
Condition 1: |R∩{Z:Z∩X≠0}|≥2 (there must be multiple groups)
Condition 2: ∀Z ∈R(Z∩X≠0 →D(X)(Z)) (the cardinal determines size of participants in group)
Importantly, note that Condition 1 looks like a plurality condition – it creates multiple groups, but plural groups do not necessarily entail plurality of events. McKercher & Kim (1999) explicitly argue that plurality of events and plurality of groups could not be the same, since they claim -ssik is compatible with single events. Thus, plurality of groups and plurality of events yield different predictions.
In order to see how these conditions apply to Serbian to get the intended meaning, we will use a simple po-sentence in (4) (a slight variation of the test sentence from Chapter 3):
(4) Plešu po tri majmuna.
dance distr three monkeys
‘distr three monkeys are dancing.’
The cardinal quantifier D in (4) is number three, X is the denotation of
monkeys and R is the denotation of the verb phrase are dancing.
Condi-tion 1, the essential function for the given semantics of -ssik/po, states that the number of sets that contain the members of the noun denotation (monkeys) and that are in the denotation of the verb phrase (are dancing) must be equal or greater than two, i.e., there should be multiple groups of monkeys dancing. Further, Condition 2 gives the number of members
5
in each group – namely, every subset of R (are dancing) that contains members of X (monkeys) must satisfy the determiner denotation of the cardinal quantifier D (three).
The crucial (and perhaps controversial) component of their semantic analysis is the claim that -ssik is not a distributive marker, it is essentially compatible with collective as well as distributive readings. We use their own examples in (5) (McKercher & Kim 1999: 248) to illustrate how their reasoning about collective readings with and without -ssik works:
(5) a. Namca sey-myeng-I kulim-ul twu-chang po-ass-ta
man three.cl.nom picture.acc two.cl look.past.dec
‘Three men looked at two pictures.’
b. Namca sey-myeng-I kulim-ul twu-chang-ssik po-ass-ta
man three.cl.nom picture.acc two.cl.distr look.past.dec
‘Three men looked at pairs of pictures.’
They show that each of these sentences has the same types of readings, logically presented in two tables (1999:249-250). Because the distributive readings are uncontroversial, we will focus only on the “collective” readings. The so-called collective reading is described as “a group of three men looked at a group of two pictures” for (5a) and “a group of three men looked at groups of two pictures” for (5b). Thus, the only thing that is different is that (5b) has multiples of pairs of pictures. Their intention of this scenario is that this is essentially one single event. To put the examples in simple illustrations, (5a) should correspond to Figure 1a and (5b) should correspond to Figure 1b:
a: Visual representation of (5a) –
collective without -ssik b: Visual representation of (5b) – collective with -ssik
Figure 1: Simple visual illustrations we created for the sentences in (5). Circles represent the
men and rectangles represent the pictures.
Before we proceed, we note that there is a conceptual problem with McKercher & Kim’s examples. Namely, the verb they chose, “to look at”, is not appropriate for the argument they are making. This verb, just like the verb “to see” is lexically distributive, meaning it carries distributive features on its own. This is often referred to as P-distributivity (in which P stands for predicate) (Winter 2001), or lexical distributivity, as discussed in Champollion (2016a). So, in the case of a “collective” scenario like the one in Figure 1b, three men could be looking individually at groups of two pictures due to the nature of the verb, and that this reading is necessarily
distributive because of the verb, resulting in multiple looking events. This, of course, can be contrasted with other verbs that are not distributive in nature, that would allow for a collective reading. These would be “mixed” verbs that are ambiguous between a collective and distributive reading, such as to carry or to pull. With such verbs, distributivity is absent on the lexical level and optional on the phrasal level (Q-distributivity) (see Winter 2001 and Champollion 2016a).
As expected, McKercher & Kim (1999) correctly predict that the Fig-ure 1a (a “true” collective reading) is incompatible with -ssik because it violates the multiple group requirement, and that that interpretation is only possible without using the marker. Similarly, po is also incompatible with Figure 1a, for a different reason – the situation violates the multiple events requirement. Note however, that both -ssik and po are compatible with Figure 1b (the problematic “collective” case) with a crucial difference being that for McKercher & Kim is a single event scenario and for Knežević this scenario involves multiple events.
A more appropriate example of these “collective” situations, using a “mixed” verb, is given by Gil (1990), who also recognizes the possible single event situation with a distributive marker -ssik in Korean and -zutsu in Japanese, but calls them NP-internal distributive readings. We present an example with this type of verb in the next section below.
2.2.2 NP-internal distributive readings
According to McKercher & Kim (1999), the “collective” readings dis-cussed in (5b) roughly correspond to the NP-internal distributive readings discussed by Gil (1990), and they refer to the latter as “double collective readings” (because both subject and direct object arguments are interpreted collectively). The “collective” term for McKercher & Kim comes from the fact these are claimed to be single event situations, while the distributive term for Gil only relates internally and does not affect the number of events. The so-called NP-internal distributive reading is a subtype of distributive readings of DistShare markers in which the relation of distributivity is ob-tained within the NP that is modified by the marker (Gil 1990).5 Consider
the examples from Gil (1990) from Japanese and Korean in (6) below:
(6) a. Otoko hutari-ga sutukeisu san-ko-zutsu-o hakonda
man two.cl.nom suitcase three.cl.distr.acc carry
b. Salam twu-myeng-i kapang sey-kay-ssik-ul wunpanha-ess-ta
man two.cl.nom suitcase three.cl.distr.acc carry.past.dec ‘Two men are carrying DISTR three suitcases.’
The general semantic interpretation of the Japanese marker -zutsu and Korean -ssik is given in (7) (Gil 1990:387). As a subtype of distributive readings, Gil defines an NP-internal distributive reading as distribution of
5 In his paper, Gil (1990) specifically remarks that the extensive analysis of Choe (1987) of Korean -ssik, although fundamentally sound, does not acknowledge this reading as distinct from distributivity over events.
5
in each group – namely, every subset of R (are dancing) that contains members of X (monkeys) must satisfy the determiner denotation of the cardinal quantifier D (three).
The crucial (and perhaps controversial) component of their semantic analysis is the claim that -ssik is not a distributive marker, it is essentially compatible with collective as well as distributive readings. We use their own examples in (5) (McKercher & Kim 1999: 248) to illustrate how their reasoning about collective readings with and without -ssik works:
(5) a. Namca sey-myeng-I kulim-ul twu-chang po-ass-ta
man three.cl.nom picture.acc two.cl look.past.dec
‘Three men looked at two pictures.’
b. Namca sey-myeng-I kulim-ul twu-chang-ssik po-ass-ta
man three.cl.nom picture.acc two.cl.distr look.past.dec
‘Three men looked at pairs of pictures.’
They show that each of these sentences has the same types of readings, logically presented in two tables (1999:249-250). Because the distributive readings are uncontroversial, we will focus only on the “collective” readings. The so-called collective reading is described as “a group of three men looked at a group of two pictures” for (5a) and “a group of three men looked at groups of two pictures” for (5b). Thus, the only thing that is different is that (5b) has multiples of pairs of pictures. Their intention of this scenario is that this is essentially one single event. To put the examples in simple illustrations, (5a) should correspond to Figure 1a and (5b) should correspond to Figure 1b:
a: Visual representation of (5a) –
collective without -ssik b: Visual representation of (5b) – collective with -ssik
Figure 1: Simple visual illustrations we created for the sentences in (5). Circles represent the
men and rectangles represent the pictures.
Before we proceed, we note that there is a conceptual problem with McKercher & Kim’s examples. Namely, the verb they chose, “to look at”, is not appropriate for the argument they are making. This verb, just like the verb “to see” is lexically distributive, meaning it carries distributive features on its own. This is often referred to as P-distributivity (in which P stands for predicate) (Winter 2001), or lexical distributivity, as discussed in Champollion (2016a). So, in the case of a “collective” scenario like the one in Figure 1b, three men could be looking individually at groups of two pictures due to the nature of the verb, and that this reading is necessarily
distributive because of the verb, resulting in multiple looking events. This, of course, can be contrasted with other verbs that are not distributive in nature, that would allow for a collective reading. These would be “mixed” verbs that are ambiguous between a collective and distributive reading, such as to carry or to pull. With such verbs, distributivity is absent on the lexical level and optional on the phrasal level (Q-distributivity) (see Winter 2001 and Champollion 2016a).
As expected, McKercher & Kim (1999) correctly predict that the Fig-ure 1a (a “true” collective reading) is incompatible with -ssik because it violates the multiple group requirement, and that that interpretation is only possible without using the marker. Similarly, po is also incompatible with Figure 1a, for a different reason – the situation violates the multiple events requirement. Note however, that both -ssik and po are compatible with Figure 1b (the problematic “collective” case) with a crucial difference being that for McKercher & Kim is a single event scenario and for Knežević this scenario involves multiple events.
A more appropriate example of these “collective” situations, using a “mixed” verb, is given by Gil (1990), who also recognizes the possible single event situation with a distributive marker -ssik in Korean and -zutsu in Japanese, but calls them NP-internal distributive readings. We present an example with this type of verb in the next section below.
2.2.2 NP-internal distributive readings
According to McKercher & Kim (1999), the “collective” readings dis-cussed in (5b) roughly correspond to the NP-internal distributive readings discussed by Gil (1990), and they refer to the latter as “double collective readings” (because both subject and direct object arguments are interpreted collectively). The “collective” term for McKercher & Kim comes from the fact these are claimed to be single event situations, while the distributive term for Gil only relates internally and does not affect the number of events. The so-called NP-internal distributive reading is a subtype of distributive readings of DistShare markers in which the relation of distributivity is ob-tained within the NP that is modified by the marker (Gil 1990).5 Consider
the examples from Gil (1990) from Japanese and Korean in (6) below:
(6) a. Otoko hutari-ga sutukeisu san-ko-zutsu-o hakonda
man two.cl.nom suitcase three.cl.distr.acc carry
b. Salam twu-myeng-i kapang sey-kay-ssik-ul wunpanha-ess-ta
man two.cl.nom suitcase three.cl.distr.acc carry.past.dec ‘Two men are carrying DISTR three suitcases.’
The general semantic interpretation of the Japanese marker -zutsu and Korean -ssik is given in (7) (Gil 1990:387). As a subtype of distributive readings, Gil defines an NP-internal distributive reading as distribution of
5 In his paper, Gil (1990) specifically remarks that the extensive analysis of Choe (1987) of Korean -ssik, although fundamentally sound, does not acknowledge this reading as distinct from distributivity over events.
5
the numeral phrase over its head noun (in which X is the numeral+classifier phrase and Y is its head noun):
(7) In every occurrence of -ssik/-zutsu, a constituent X containing -ssik/-zutsu distributes over a semantically plural constituent Y disjoint from X: [X … ssik/zutsu … ] distributes over [Y[+pl] … ]
Gil differentiates three main types of distributive readings for (6) (and by extension, for (5b)) – an individual-distributive (participant) distributive reading in which there are two men and six suitcases in total (three suitcases per man), resulting in two events of carrying; an event-distributive reading in which there are multiples of two men carrying triples of suitcases (at least four men carrying at least six suitcases – three suitcases per two men), resulting in at least two events of carrying; and a NP-internal distributive reading in which there is a subtle, although crucial difference from distrib-utive readings over events (aka over the verb). Specifically, distributivity over events necessarily entails a plurality of events, while there is no such entailment for NP-internal distributive readings. This means that, for Gil, distributive readings with -ssik (and po by extension) are compatible with single-event scenarios. For McKercher & Kim (1999), these readings are possible but they highlight that -ssik does not contribute to distributivity, but only to the plurality of groups constrained by the cardinality.6
In (6), NP-internal distributive reading is obtained via distributing the numP DISTR three (sets of three) over the NP suitcases and it is illustrated as a situation in which, for example, two men are together struggling to carry a huge sack of 15 suitcases which are organized in five batches, each batch with three suitcases. If this was an event-distributive reading, this situation would require five events of carrying groups of three suitcases. So, the situation described above may be considered as a type of “collective” situation – it involves one and only one event of agents collectively carrying suitcases grouped in batches of three. The predicate itself is thus not being interpreted distributively. The fact that the agents here are acting collectively and the theme (the suitcases) is collectively acted upon, leads McKercher & Kim to refer to these readings as “double collective”. The supposed availability of these readings also gives them further support that the marker in Korean (and Japanese) is perhaps better seen as a group-forming device than a distributive marker.7
2.2.3 Counterarguments to the group-forming analysis
However, we offer here counterarguments to the assumption that DistShare markers are simply a group-forming device that can allow single-event
6 This is a point of divergence between McKercher & Kim (1999) and Gil (1990) – Gil still ascribes the distributive property to the marker, while McKercher & Kim do not and claim distributivity comes from elsewhere.
7 Note, however, that purely distributive verbs in collective scenarios with the marker po (or -ssik) (e.g.,
eat, drink, see) create a problem in distinguishing single events vs. multiple distributive events (i.e.,
one cannot drink or see something collectively, only at the same time), as discussed in 2.2.1, which is not something McKercher & Kim address at all.
readings. The first problem occurs with examples in which we should, but are not able to, generate NP-distributive readings that involve just one single event, such as (8):
(8) Jovan je izmerio po tri jabuke.
John aux weighed.telic distr three apples.acc
‘John weighed distr three apples.’
If we follow McKercher & Kim’s (or Gil’s) analysis, and assume po, just like -ssik, is a group-former, we should be able to generate a scenario where there is a single-event, NP-internal distributive reading/ double collective reading, for (8). Therefore, we should have Jovan, in a single event of weighing, weigh multiples of three apples on one scale in a way that the theme (the total number of apples) is collectively acted upon by the agent (Jovan). However, the plausibility of such scenarios makes little sense – there is no pragmatic necessity for differentiating multiples of three apples when the result is not distinguishable from a true collective reading that can be simply obtained by mentioning the total number of apples. More precisely, a single scale does not distinguish the weight of groups of three apples.
We then have a problem of over-generating with the group-forming hypothesis. The same over-generating problem appears especially with telic/accomplishment verbs that can refer to a single event (e.g., cook (up),
write (up), wash (up), etc). Consider the example in (9):
(9) Maja je skuvala po pet krompira.
Maja aux cooked.up.telic distr five potatoes.acc
‘Maja cooked (up) distr five potatoes.’
For a double collective reading we should have Maja, in a single event of cooking, boil and cook up multiples of five potatoes in a single pot. Here too the scenario makes no sense because the end result is indistinguish-able from a collective reading – a single pot cannot differentiate between cooking groups of five potatoes. Unlike it is the case for (8), we could potentially imagine a scenario for (9) in which the pot has separators and it allows potatoes to be cooked on different levels of the pot. If the levels contain groups of five potatoes, perhaps it would be possible to truthfully say (9). Note that if the same context is applied to (8) we would end up with multiple scales and then the question still remains how to differentiate between the weights of different triples of apples.
For a distributive scenario to work (and make sense especially for (8)), we should also require multiples of the instrument/location argument – multiple scales and multiple pots. This, however, requires the multiplication of an argument that is not linguistically expressed (and even if it is, it is not marked by the marker po). Thus, NP-internal distributive readings are hard to achieve unless there is also some additional parameter that must vary and results in some form of distribution.
5
the numeral phrase over its head noun (in which X is the numeral+classifier phrase and Y is its head noun):
(7) In every occurrence of -ssik/-zutsu, a constituent X containing -ssik/-zutsu distributes over a semantically plural constituent Y disjoint from X: [X … ssik/zutsu … ] distributes over [Y[+pl] … ]
Gil differentiates three main types of distributive readings for (6) (and by extension, for (5b)) – an individual-distributive (participant) distributive reading in which there are two men and six suitcases in total (three suitcases per man), resulting in two events of carrying; an event-distributive reading in which there are multiples of two men carrying triples of suitcases (at least four men carrying at least six suitcases – three suitcases per two men), resulting in at least two events of carrying; and a NP-internal distributive reading in which there is a subtle, although crucial difference from distrib-utive readings over events (aka over the verb). Specifically, distributivity over events necessarily entails a plurality of events, while there is no such entailment for NP-internal distributive readings. This means that, for Gil, distributive readings with -ssik (and po by extension) are compatible with single-event scenarios. For McKercher & Kim (1999), these readings are possible but they highlight that -ssik does not contribute to distributivity, but only to the plurality of groups constrained by the cardinality.6
In (6), NP-internal distributive reading is obtained via distributing the numP DISTR three (sets of three) over the NP suitcases and it is illustrated as a situation in which, for example, two men are together struggling to carry a huge sack of 15 suitcases which are organized in five batches, each batch with three suitcases. If this was an event-distributive reading, this situation would require five events of carrying groups of three suitcases. So, the situation described above may be considered as a type of “collective” situation – it involves one and only one event of agents collectively carrying suitcases grouped in batches of three. The predicate itself is thus not being interpreted distributively. The fact that the agents here are acting collectively and the theme (the suitcases) is collectively acted upon, leads McKercher & Kim to refer to these readings as “double collective”. The supposed availability of these readings also gives them further support that the marker in Korean (and Japanese) is perhaps better seen as a group-forming device than a distributive marker.7
2.2.3 Counterarguments to the group-forming analysis
However, we offer here counterarguments to the assumption that DistShare markers are simply a group-forming device that can allow single-event
6 This is a point of divergence between McKercher & Kim (1999) and Gil (1990) – Gil still ascribes the distributive property to the marker, while McKercher & Kim do not and claim distributivity comes from elsewhere.
7 Note, however, that purely distributive verbs in collective scenarios with the marker po (or -ssik) (e.g.,
eat, drink, see) create a problem in distinguishing single events vs. multiple distributive events (i.e.,
one cannot drink or see something collectively, only at the same time), as discussed in 2.2.1, which is not something McKercher & Kim address at all.
readings. The first problem occurs with examples in which we should, but are not able to, generate NP-distributive readings that involve just one single event, such as (8):
(8) Jovan je izmerio po tri jabuke.
John aux weighed.telic distr three apples.acc
‘John weighed distr three apples.’
If we follow McKercher & Kim’s (or Gil’s) analysis, and assume po, just like -ssik, is a group-former, we should be able to generate a scenario where there is a single-event, NP-internal distributive reading/ double collective reading, for (8). Therefore, we should have Jovan, in a single event of weighing, weigh multiples of three apples on one scale in a way that the theme (the total number of apples) is collectively acted upon by the agent (Jovan). However, the plausibility of such scenarios makes little sense – there is no pragmatic necessity for differentiating multiples of three apples when the result is not distinguishable from a true collective reading that can be simply obtained by mentioning the total number of apples. More precisely, a single scale does not distinguish the weight of groups of three apples.
We then have a problem of over-generating with the group-forming hypothesis. The same over-generating problem appears especially with telic/accomplishment verbs that can refer to a single event (e.g., cook (up),
write (up), wash (up), etc). Consider the example in (9):
(9) Maja je skuvala po pet krompira.
Maja aux cooked.up.telic distr five potatoes.acc
‘Maja cooked (up) distr five potatoes.’
For a double collective reading we should have Maja, in a single event of cooking, boil and cook up multiples of five potatoes in a single pot. Here too the scenario makes no sense because the end result is indistinguish-able from a collective reading – a single pot cannot differentiate between cooking groups of five potatoes. Unlike it is the case for (8), we could potentially imagine a scenario for (9) in which the pot has separators and it allows potatoes to be cooked on different levels of the pot. If the levels contain groups of five potatoes, perhaps it would be possible to truthfully say (9). Note that if the same context is applied to (8) we would end up with multiple scales and then the question still remains how to differentiate between the weights of different triples of apples.
For a distributive scenario to work (and make sense especially for (8)), we should also require multiples of the instrument/location argument – multiple scales and multiple pots. This, however, requires the multiplication of an argument that is not linguistically expressed (and even if it is, it is not marked by the marker po). Thus, NP-internal distributive readings are hard to achieve unless there is also some additional parameter that must vary and results in some form of distribution.
