3.1.1 Landman (2016)

According to Landman, the internal structure of an IHRC is as follows:

The trace t_n is interpreted as an individual variable x_n, abstracted over ultimately by Op_n. The phonetically null P is stipulated to pick out a thematic role π among the roles associated with the event predicated by the head X of XP and to identify the bearer of π with x_n. If we take XP₂ to denote an event predicate E, the interpretation of XP₁ is:

Existential closure of the event variable occurs below the predicate abstraction induced by Op_n. In the case in which the IHRC contains only a single clause with no embedding, the interpretation for the CP as a whole can be simplified as follows, ignoring tense and aspect:

Since π is selected from among the roles assigned within E, π(e) ends up doubly restricted in (17). This is best seen by looking at an example. Consider the IHRC Ayaka-ga kukkii-o tsukut-ta-no (= Ayaka made cookies). With the IH coming from the Th(eme) role associated with kukkii, the CP has the following (simplified) interpretation:

The CP interpretation derived in this manner is a property denotation. Landman assumes “that the implicit definiteness operation σ [where σ(P) is the sum of the elements in P on the presupposition that that sum itself is in P] brings the relative in argument position from a predicative meaning to a definite interpretation at the type of individuals.” Concretely, this operation generates (19) as the final interpretation of the IHRC.

This denotes the maximal plural individual that is cookies that Ayaka made.

Landman also incorporates a formal analysis of K(uroda) R(elevency) effects into the semantics of IHRCs. Intuitively, KR effects indicate a restriction on the relation that can hold between the matrix event of which an IHRC is an argument on the one hand and the IHRC event on the other. Landman builds this restriction into the semantics of the empty postposition P used to select a thematic role for identifying the IH of the IHRC. He does so by employing a contextually supplied Kuroda Relevancy function k, where for all e, e′, k(e)=e′ iff e, e′ satisfy the KRC (= (9)). Incorporating this function into the semantics of P results in the following schematic interpretations for XP₁ and CP in (15), where e′ is a free variable:

After application of the definiteness operation σ, e′ is ultimately identified with the event variable of the matrix event predicate, establishing the desired connection between the IHRC event and the matrix event.

While Landman’s analysis represents clear progress in analyzing IHRCs and accounting for their KR effects, the analysis faces two problems. First, it is central to Landman’s analysis that free variables be allowed to take on contextual values, as seen in the fact that the free Kuroda Relevancy function variable k does so. Allowing the same possibility for the free event variable introduced in (20), however, causes problems. In particular, it leads to the incorrect prediction that KR-violating IHRCs can be salvaged by embedding them in a discourse with an antecedent that together with the IHRC satisfies the KRC. For example, (21) is predicted to be acceptable, contrary to observation.

The second sentence of (21) is unacceptable on its own as an IHRC, presumably because the events e of Ayaka’s making cookies and e′ of the cookies being chocolate chip cookies do not satisfy the KRC.¹⁰ The following sentence, however, is acceptable, indicating that the KRC is satisfied by e and the event e″ of Taro’s putting the cookies in the fridge.

Since the first sentence of (21) is nothing more than an independent characterization of this e″, if e″ could be identified pragmatically as the value of k(e) within the IHRC in the second sentence of (21), the KRC would be satisfied and the example predicted to be acceptable. Somehow this result needs to be blocked, but Landman gives us no principled basis for doing so.

The second problem with Landman’s analysis is that even restricting the relevant events to sentence-internal events, it makes incorrect predictions about which events need to satisfy the KRC. Since the KR operator is introduced by the P that locally fixes the IH as a thematic participant in the XP₂ sister of its PP projection in (15), the analysis predicts that KR should be required to hold between the main XP₂ event and an IHRC-external event. For IHRCs with only a single clause this prediction is borne out. For IHRCs whose IH resides in an embedded clause, however, it is not. This can be seen by comparing the following examples.

The acceptability of (23a) shows that the KRC is satisfied by the IHRC event of Ayaka making cookies and the matrix event of Junya eating them. In (23b), if the KRC were required to hold between the most embedded event and the matrix event, the example would be predicted to be acceptable since this results in the same two events having to satisfy the KRC as were seen to satisfy it in (23a). The fact that (23b) is not acceptable strongly suggests that the KRC relates the matrix event not to the event of the minimal IH-containing clause but to the event of the highest clause in the IHRC. The example in (23c) shows that the corresponding EHRC, which lacks any KRC requirement, is acceptable, further corroborating this suggestion.

The problem of identifying the events that the KRC constrains shows up not only with the IHRC-internal event but also with the IHRC-external event. Kuroda (1999a; b) identifies this latter event as the main event of the expression that selects the IHRC. While Landman’s analysis allows this to be the relevant event, since it relies essentially on binding of a free event variable e′, and since variable binding can occur over arbitrarily long distances, Landman’s analysis predicts that it should be possible for an example to satisfy the KRC non-locally. In particular, an IHRC selected by a head H with respect to which the KRC is not satisfied is predicted to be salvageable if the entire [IHRC H] construction is embedded under a higher predicate H* when the KRC is satisfied by the IHRC and H*. This prediction as well is not borne out, as seen in the examples below.

(24a) shows again that the eventualities of Ayaka making cookies and of the cookies being chocolate chip cookies do not satisfy the KRC. (24b) shows that the events of Ayaka making cookies and of Junya enjoying them do satisfy the KRC. If the KRC could be satisfied long distance, (24c) would be expected to be acceptable since it would essentially relate the same two events as (24b). That it is not acceptable where its EHRC counterpart in (24d) is acceptable strongly suggests that the problem with (24c) stems from its not satisfying the KRC.

3.1.2 IHRC syntax

We analyze IHRCs syntactically as bare CPs, lacking an EH. We introduce an IHRC-specific operator Op that exploits the mechanisms of thematic role assignment to identify the IH, replacing the ad hoc ChR operator from Grosu & Landman (2012) and the equally ad hoc empty postposition P from Landman (2016). We follow Champollion (2015) in separating lexical predicates from thematic role assignment both in the syntax and in the semantics.¹² We take thematic role assigners to be syntactic expressions that combine first with an argument of a syntactic predicate and then with the predicate itself. We accomplish IH-identification by adding a second occurrence of a particular thematic role assigner to a predicate. The internal syntactic argument of this thematic role assigner is an operator that raises eventually to SpecCP, where it takes three arguments: (i) (the predicate abstraction of) its C′ sister, (ii) an IHRC-external thematic role assigner, and (iii) an IHRC-external event quantifier (typically a projection of the matrix V). This raising is assumed to give rise to island effects. The schematic internal structure of the CP of an IHRC is given below, where θ is a thematic role assigner that assigns a role of the head of X₂ to an individual. Note that only the IHRC-internal argument of Op in (i) is represented, not the IHRC-external arguments in (ii) and (iii).¹³

Before developing the analysis further, a word is in order regarding our syntactic assumptions. The analysis we are adopting for IHRCs shares with Kuroda (1975/76; 1992; 1999a; b) the idea that an IHRC is a nominalized sentence without any external head, with no as the nominalizer (for Kuroda, analyzed as a nominalizing complementizer). For us, an IHRC is a CP to which no is adjoined. It is a common complaint against such analyses that CPs cannot generally occur in Case-marked positions (Stowell 1981). We analyze the occurrence of no that obligatorily accompanies IHRCs as a Case-receiver, a semantically vacuous expression¹⁴ bearing the feature [+N], whose sole function is the syntactic function of receiving Case from some external Case assigner. In this respect, our analysis of no in IHRCs is similar to the standard analysis of English of. In nominalizations like destruction of the city, the preposition of is standardly taken to assign Case to the object. It shows up for purely syntactic reasons, having no identifiable semantic content. The difference between our analysis of the occurrence of no that marks IHRCs and the standard analysis of of lies in what syntactic role the two expressions play: of assigns case, while no receives case. In this sense the two can be thought of as complementary particles.¹⁵ What remains is to give the semantic interpretations needed to get these analyses to work.

3.1.3 Semantics

We base our formal analysis on the event semantics of Champollion (2015).

3.1.4 Explanations

The necessary movement of Op from a position in a sister to the minimal clause containing the IH to the highest position in the IHRC accounts for the island sensitivity of IHRCs in exactly the same way as Landman’s (2016) analysis. Definiteness, as highlighted earlier, is accounted for through the stipulated inclusion of Landman’s σ operator in the interpretation of Op. KRC effects come from explicitly requiring two events in the semantics of Op to satisfy a KR relation, another stipulation of the analysis. Which two events get related derives from the requirement that the events be the main events of two of the semantic arguments of Op. This makes it impossible for the IHRC-internal event to be anything but the main event of the sister of Op, and makes it impossible for the IHRC-external event to be anything but the event of the predicate that selects the IHRC.

Regarding integration, as noted in section 1, we take a construction to exhibit integration if it may be contained within a single illocutionary unit. Non-restrictive relative clauses do not meet this requirement, but IHRCs, as analyzed above, do. The compositionally derived semantic interpretation of (36) given in (38), for example, can easily be uttered as a single assertion, questioned in a single question, or have any part of it negated with the addition of a single negation.

As the analysis stands, it does not block IHRCs with individual-identifying IHs. We achieve the desired blocking by only allowing a single identifying thematic role assigner per event quantifier. Identifying thematic role assigners, recall, add conjuncts of the form θ(e)=x to an event quantifier, not merely restricting θ(e) like restrictive thematic role assigners do but identifying a specific individual referent for θ(e). Blocking event quantifiers from combining with multiple occurrences of a single identifying thematic role assigner θ_i makes it impossible to add an occurrence of [Op θ_i] to an event quantifier that has already combined with some other occurrence of the form [x θ_i], which thereby makes it impossible to generate an IHRC with an individual-identifying IH.

For definite descriptions, we distinguish those associated with a known referent from those merely known to have a unique referent. The former we analyze as of type e or ⟨et,t⟩ and relate to an event via an identifying thematic role. The latter we analyze as of type et and relate to an event via a relational thematic role. This makes it possible to generate an IHRC with a definite description as IH whenever that description does not have a known referent, a possibility illustrated in the following sentence.

We can know that there is a unique densest planet without knowing which planet that is. The description uchuu-de mottomo mitsudo-ga takai wakusei (the densest planet in the universe) can then be analyzed as a type et expression and its denotation can be entered into the event structure of chikazuitekita (get closer) with a relational thematic role assigner. This makes it possible to add a separate identifying role assigner to the same predicate, rendering it possible to generate an IHRC with this description as IH.

3.2.1 Erlewine & Gould (2016)

EG attempt to unify EHRCs, IHRCs and DHRCs. They analyze the three constructions as syntactically derived from a single underlying representation, and as differing only (i) in which of two operations applies, Trace Conversion (TC) or Inverse Trace Conversion (ITC), and (ii) in whether the relevant operation applies before or after Spell-Out. The appeal of such an approach is obvious: it aims to account for the properties of all three constructions using only independently available tools. The challenge such an approach faces is in accounting for systematic differences among the three constructions, and – as already pointed out by Grosu & Hoshi (2018) – in providing an analysis for all the varieties of EHRCs, IHRCs and DHRCs found in Japanese.

For EG, all three relative clause constructions are constructed from a proto relative clause, derived by copying an RC-internal expression and merging the RC with the NP inside that copied expression. For a relative clause based on the sentence Ayaka-ga kukkii-o 20-mai tsukut-ta (= Ayaka made 20 cookies), their proto relative clause structure is the following:

One of two operations can apply to (41): TC or ITC. TC adds a variable x to the RC-internal kukkii, equates x with the reference of kukkii, converts everything in the lower DP copy that is not part of the NP into the determiner THE, and binds x with λ_i, as in (42a). ITC adds a variable X to the RC-internal kukkii, analyzes the reference of kukkii as a part of X, binds X with λ_i, and converts everything in the higher copy that is not part of the NP into the determiner THE, as in (42b).²¹

TC and ITC can each apply either before or after Spell-Out. If they apply after Spell-Out, the structure that gets spelled out is that in (41). EG take the RC-internal copy and the RC-external copy of kukkii 20-mai to be formally the same object in (41), which can only be spelled out in one position. If this object is spelled out RC-internally the result is an IHRC, and if spelled out RC-externally the result is an EHRC. EG additionally stipulate that the position in which this object is spelled out at PF is the position in which the quantificational material is interpreted at LF.²² This stipulation essentially requires application of TC in the EHRC case, generating the LF representation in (42a), and application of ITC in the IHRC case, generating the LF representation in (42b).

When TC and ITC apply before Spell-Out, the structures that get spelled out are those in (42). In these cases, the two copies of kukkii presumably do not count formally as the same object, making it possible for both copies to be pronounced without violating linearization requirements. The combination of THE with kukkii in (42b) is taken to be conventionally spelled out as sono kukkii. The parallel combination in (42a) is stipulated to not be pronounced at all.

Regarding the KRC, EG stipulate that “pronunciation of the lower copy nominal – the property shared by [IHRCs] and DHRCs, to the exclusion of [EHRCs] – is licensed by satisfaction of the Relevancy Condition.” They thus predict that KRC effects should arise in both of these constructions.

3.2.2 Challenges for Erlewine & Gould (2016)

If EG’s analysis were empirically adequate, the many stipulations needed to make it work could be accepted as a necessary cost for success. The analysis is not, however, empirically adequate. In this section we outline empirical problems for EG’s analysis that any analysis of DHRCs must explain.

3.2.3 DHRC syntax

While superficially similar to IHRCs in having a gapless relative clause, our formal analyses of DHRCs and IHRCs are very different. First, DHRCs have an overt EH, while we analyzed IHRCs as lacking an EH. Second, we analyze DHRCs as grammatizations of E-type anaphora. Like all E-type anaphora, we take the denotation of a DHRC to be recovered from an event rather than from an individual. In contrast to discourse anaphora, the event from which a DHRC recovers its reference lies not in a thematically independent antecedent but rather in its RC argument.

We analyze DHRCs syntactically as gapless relative clause constructions whose EH is a so- expression.²³ When the EH is of the form sono NP it can be understood either as a partitive or as a non-partitive. As a partitive, we analyze the DHRC as a whole as consisting of a DHRC proper having the form [RC so+proN], combined with a genitive case marker and a part-denoting NP, structured as in (50a), where proN is a phonetically null nominal of type et. As a non-partitive, we take so- to combine directly with the overt NP, and for sono NP as a whole to be the EH of the DHRC proper, as in (50b). When the EH consists of a single word such as soitsu, sore, etc., there is no partitive/non-partitive ambiguity, and we analyze the DHRC as having a structure parallel to (50c).

3.2.4 DHRC semantics

We analyze the relative clause of a DHRC as a normal clause, without closure (like all integrated embedded clauses). This means it has the semantics of an event quantifier, of type ⟨vt,t⟩. The important work of generating a referent from such a clause is done by the semantic head so-, interpreted as follows:

So- takes a nominal property N and an event quantifier V and gives back an individual. The individual denoted is the maximal individual x such that x has the property N and x is a participant of the main event of V. An individual x is a participant in an event e iff e wouldn’t exist without x.²⁴ For example, an event of Ayaka making cookies wouldn’t exist without Ayaka or without the cookies she makes, and so both Ayaka and the cookies she makes are participants in that event. We illustrate the analysis below:

3.2.5 Explanations

Under the proposed analysis of DHRCs, the RC does not introduce an independent illocutionary act, instead serving to restrict the reference of the DHRC as a whole. This renders DHRCs integrated. This integration derives from the semantics of so-, which allows it to take an event quantifier-denoting clause as a semantic argument. Since the otherwise similar a- and ko- do not have interpretations that allow them to do so, they cannot form integrated DHRCs in the same way.²⁵

Our analysis predicts that a DHRC cannot have a unique IH that is contained within a non-integrated clause. The prediction comes from the interaction of the participant relation and integration. For an integrated event quantifier interpretation E, the participants of the main event e of E are those individuals that directly participate in e, or in event arguments of E, or in event arguments of event arguments of E, etc. Non-integration breaks this chain of connection. A clause C that is not integrated with E is a clause that is not connected to E via any such chain, blocking participants in events in C from being participants of e. The only exception to this is cases where a participant of an event of C and a participant of e are independently represented and identified with each other, e.g. via anaphora. In the absence of such an anaphoric connection, the participants of events in C and the participants in e are non-overlapping. From this fact we get the prediction that for a DHRC with a unique IH, its IH cannot be in a sub-CLause of the DHRC that is not integrated with the main predicate of the relative clause. That this prediction is borne out can be seen by comparing the following examples, where the non-integrated clause is underlined.

(53a) shows that there is no problem in principle with constructing a DHRC from an RC that contains a non-restrictive clause. We see in (53b), however, that that non-restrictive clause cannot contain the head of the DHRC. For analyses of DHRCs like that of Kitagawa (2019) that take the EH to be anaphoric on a referential antecedent, the unacceptability of (53b) is hard to explain.

The insensitivity of DHRCs to islands derives from the fact that identification of the IH does not involve movement. The IH is identified as the unique individual that satisfies so-’s internal argument and that is a participant of the main event of the DHRC’s relative clause. The participant relation is a semantic-pragmatic relation that is potentially non-local.

A word is in order about the nature of the participant relation. In order to account for the acceptability of pseudo-DHRCs like (46), we have to assume that a smell (nioi) is a participant of an event of a fish getting grilled (sakana-ga yake-ta). Clearly there is no logical connection between events of fish getting grilled and their associated smells. It is logically possible to grill a fish without generating any odor whatsoever. However, there is a clear contextually salient connection between the two: in the real world, grilling a fish generates an odor, and would arguably not be the same event if it did not do so. This means that the participant relation has to be analyzed as at least potentially involving contextually salient connections between individuals and events rather than logically necessary connections between the two.

The definiteness of DHRCs comes from the σ operator contained in the lexical semantics of their semantic head so-. The denotation of a DHRC is the denotation of a logical expression of the form σ(P), which by the definition of σ gives us the unique maximal individual that P is true of.

The lack of KRC effects in DHRCs comes from the absence of a KR operator in the semantics of so-. Formally such an operator could be added to so- just as it was included in Op, meaning that the absence of KRC effects in DHRCs is not a theoretical necessity any more than their presence is a necessity in IHRCs. The fact that our analysis of DHRCs differs from our analysis of IHRCs in not involving Op makes it unproblematic, however, that the two constructions differ with respect to KRC effects.

The IH of a DHRC can be an identifying expression such as a name or a demonstrative, setting DHRCs apart from IHRCs. The explanation in section 3.1.4 for the impossibility of individual-identifying IHs in IHRCs rests on the assumption that individual-identifying expressions are of type e and cannot be type-shifted, thus having to be entered into an event domain using an identifying thematic role assigner. This assumption that individual-identifying expressions of type e cannot be type-shifted results in a type clash when combining so- in (51) with a name or demonstrative in the EH position of a DHRC. We overcome this problem by allowing so- to type shift to the following:

Since IHRCs are not semantically headed by so-, this addition has no effect on their interpretation.

We saw in section 3.2.2, in particular in examples (43) and (45), that a DHRC with the quantificational IH hotondo-no kukkii (= almost all of the cookies) is restricted in its possible interpretation. If we distinguish the domain over which hotondo operates – the 20 cookies Ayaka made – from the witnesses that render the relative clause true – the 18 cookies that Yoko put in the fridge – in those examples, we saw that for a DHRC the denotation can be the witnesses but cannot be all of the cookies in the domain unless they are all witnesses as well. On our analysis, this fact is derived from the interaction of the semantics of hotondo with the participant relation in the semantics of so-.

We analyze hotondo as operating over a contextually identified domain D. For hotondo-no kukkii, we take D to be a plurality of cookies, i.e. a sum of atomic cookies (cf. Link 1983). We analyze hotondo as follows, where ⊑ is the (plural) part-of relation, D-x is the plural individual that results from removing all the atomic individuals in x from D, |x| is the number of atomic individuals that are part of x, and m≫n is true iff m is far greater than n.

This produces the following interpretation for the DHRC in (43):

This denotes the cookies that are participants of some event of Yoko putting cookies into the fridge. The important thing to focus on here is the different roles played by D and z with respect to the putting event e′. In particular, a witness for z is directly thematically related to e′ in the final line, making that witness a participant of e′. In contrast, D, the domain of cookies, only serves to place restrictions on z. It is not thematically related to e′ either directly or indirectly and does not count as an independent participant of e′ in any other way. The denotation of the DHRC is then the maximal plurality x of cookies such that for some event e′ of Yoko putting cookies in the fridge, x is a participant of e′. For a situation like the one envisioned for (43) in which there is only a single event of Yoko putting cookies in the fridge, this results in the DHRC denoting the witnesses for z, not the domain D, with respect to that event.

We saw in section 3.2.2 (Mismatched IH and EH) that the EH and the IH of a DHRC can be mismatched. This is expected under our analysis. Consider (48), repeated here.²⁶

This example has a single EH, sono kudamono (that fruit), but two corresponding IHs, san-ko-no nashi (three pears) and ni-ko-no ringo (two apples). According to the semantics of so-, the DHRC is predicted to refer to the maximal plurality of fruit that counts as a participant in the conjoined buying event, i.e. to the three pears and two apples that John and Mary bought. This is exactly what the DHRC is understood to refer to.