Mechanisms of productivity in word formation: Transitivity alternations in Japanese

The extensive system of Japanese transitivity alternations (TrAlt) has inspired a great number of noteworthy works. This paper, adopting the theoretical apparatus of the Strong Minimalist thesis and focusing on the “ in-between voice” (i.e., between the idiosyncratic lexical and the morpho-syntactic) characteristics of the TrAlt, aims to explicate its productivity phenomena. The paper offers detailed and novel analyses of transitivity suffixes -ar (a lexical passive) and -e (which has sometimes been characterized as a “causative-passive” suffix), and delineates the mechanisms of productivity by closely examining the interactive functions of these two suffixes and the verbal roots (R’s) that host them. CHISATO KITAGAWA Mechanisms of productivity in word formation: Transitivity alternations in Japanese CORRESPONDING AUTHOR: Chisato Kitagawa University of Massachusetts Amherst, US


Introduction
The extensive transitivity alternation system in Japanese (henceforth TrAlt) 1 is represented by doublet pairs composed, in each case, of a verbal root (henceforth R) and a transitivity suffix signaling the transitivity opposition. Its defining characteristic, as pointed out by Noda (1961:214), is its "in-between voice" status (tyuukan no voice in his words). 2 The TrAlt represents the "in-between voice," because it operates somewhere between the truly idiosyncratic lexical formation expressing voice distinction and the syntactic operation involving (r)are direct passivization and periphrastic (s)ase causativization. This "in-between voice" characteristically manifests in seemingly contradictory phenomena in that, on one hand, the transitivity suffix that a particular R takes is "not predictable" (Hasegawa 2001: 2) but, on the other, the TrAlt also allows, as pointed out by Nishio (1954), Miyaji (1985), and Kageyama (1996), quite robust zoogo 'new word creation' (Kageyama 1996: 191) ("new" in the sense that the given formations are not registered in earlier dictionaries). 3 In this respect, Japanese TrAlt presents a picture significantly different from the voice alternation systems of European languages discussed in Lavine and Babby (2019). It also is an issue not covered in the extensive cross-linguistic study of transitivity alternation by Alexiadou et al. (2015).
My objective is to delineate the mechanisms of this productivity, restricting my attention to the following three transitivity suffixes, -ar (henceforth AR), -as (henceforth AS), and -e (henceforth E). 4 To that end, I adopt the theoretical apparatus provided by the Strong Minimalist thesis of Chomsky (2013;2015), especially in a way articulated in Epstein et al. (2016), 5 in order to provide a novel explication of the precise functions of these "transitivity" suffixes, in particular AR and E. The exact functions of these suffixes with respect to the productivity phenomena have never been sufficiently identified in the literature in spite of the fact that the TrAlt has been subject to intense studies for a long period of time. Without clarification of such matters, no study of Japanese TrAlt properly contributes to the cross-linguistic understanding of word formation involving voice alternations.
The outline of the paper is as follows. Section 2 provides the general characteristics of Japanese TrAlt. Section 3 offers detailed analyses of the two distinct ways in which suffix AR can manifest -either as "lexical passive" or as simple "intransitivizer." Section 4 discusses the productivity phenomena involving AR. Section 5 is a brief account of lexical causative AS. 6 In Section 6, I present a unified and novel analysis of multi-faceted manifestations of suffix E, highlighting its peculiar behavioral patterns intricately interacting with those of lexical passive 1 The range of TrAlt pairs is broader, for instance, than the corresponding system in English described in Levin and Rappaport Hovav (1995) (see Matsumoto 2016c: 54).

3
For example, an internet blog site, hiiragi-june 'holly-June' dated 2007-09-05, commented on a newly created transitive verb [tati.ag-E] 'start out (a project)' first appearing as a computer related new word (e.g., pasokon-o tati. ag-e 'start up a personal computer') formed on the basis of the already existing intransitive verb [tati.ag-AR] 'stand up, get up' by analogy to one of the doublet patterns in the TrAlt system (to be identified as Type (1C) {[R-AR] intr / [R-E] tr }). The use of [tati.ag-E] tr later expanded beyond the sphere of computer discourse: e.g., Venture kigyoo-o tati. ag-E 'start up a venture enterprise'. The authoritative and conservative Japanese language dictionary Koojien listed this transitive tati.ag-E for the first time in its Fifth Edition published in November 1998 as a computer related word, and a more trend-sensitive Japanese language dictionary Iwanami Kokugo Jiten 'Iwanami Japanese Language Dictionary' listed it for the first time in its Fifth Edition published in November 1994. We will return to this "new word formation by analogy" issue in Section 7.

4
In this, I emulate Kageyama's (1996) practice. It is also instructive that Narrog (2016: 257), in his detailed diachronic study, cites these three as the representative formatives contributing to the TrAlt system.

5
This framework, as recognized by Epstein et al. (2016: 89), is similar to that of Marantz's Distributed Morphology; to quote Marantz (1997: 205): "To imagine a theory in which the grammar constructs all words in the syntax by the same general mechanisms ("merge and move," see Chomsky 1995) that construct phrases, it is useful to make the natural assumption that whether you get a "zero-level category" (word-like unit) or a phrasal category by merging two constituents is a function of the constituents involved, not of the "merger" operation itself. That is, there is no reason not to build words in the syntax via "merger" (simple binary combination) as long as there are no special principles of composition that separate the combining of words into phrases from the combining of morphemes into words."

6
For the extensive analysis of AS (along with that of E and AR), see Matsumoto (2016c: 53), which presents a detailed study of the Japanese TrAlt, utilizing "a hierarchy of schemas such as those proposed in Booij (2007;2010) in the framework of Construction Morphology," a theoretical framework quite different from that adopted by such authors as Comrie (2006), Haspelmath (1993), and Jacobsen (1992). Matsumoto's analysis unearths some intriguing, and hitherto unknown, phonological subregularity maintained in the Japanese TrAlt system.

Japanese transitivity alternation (TrAlt)
TrAlt is a strikingly familiar phenomenon in Japanese. The following chart is based on Jacobsen's (1992: 258-268) "Appendix" which "contains all transitive and intransitive verbs falling into pairs which are listed in Kenkyuusha's New School Japanese-English Diectionary (1968)" with "some pairs added from Shimada 1979" (Jacobsen 1992. 7 In (1), the suffix pattern {X/Y} indicates that, given the identical R, the R-X represents an [R-X] intr and the R-Y an [R-Y] tr , constituting thus a transitivity doublet pair. The symbol "∅" stands for "phonological null." (1) Type Suffix Pattern Intransitive, e.g. Transitive, e.g.
TrAlt is thus marked "by a series of verbal oppositions cutting across the native verb system" (Jacobsen 1992: 56); all participating verb pairs can be said to represent a changeable state of the subject/object, and therefore can be said to have "a theme or theme-like subject/object" (Matsumoto 2016: 53). This "opposition" is expressed by transitivity suffixes AR, AS, and E.
Following the spirit of Marantz (1984), I assume that those suffixes are variants of the category v, which, as noted by Pylkkänen (2008: 7), stands for any functional head that is of the verbal category. 8 With Chomsky (1995;2015), v stands for the universal functional category directly involved in the building of predicates. For the purpose of this paper, I venture to depart somewhat from Chomsky's (2013: 43, fn. 29) specific characterization of v that it has "(at least) two ''flavors'': v for unaccusative/passive, v* for transitive/unergative. The reason for this departure is that Japanese is an agglutinative language. Of agglutination, Poppe (1965: 190) comments: "All Altaic languages are agglutinative in structure. Agglutinative means that inflection and word formation take place by adding suffixes to stems" (Poppe 1965: 180); "the suffixes are monofunctional, contrary to the polyfunctional endings of inflectional Indo-European languages." Accordingly, I claim that the Japanese v suffixes AR, AS, and E are all monofunctional. The purpose of this paper is thus to define the nature of the monofunctionality undertaken by each of these suffixes. In this scheme, such properties as unaccusativity or unergativity belong to the R categories. Translating Chomsky's above specification for the present paper, I would only assume that v has "(at least) two flavors'': v for passive-orientation, and v* for transitive-orientation; I use "v" hereafter as the representative term, however, unless the distinction becomes crucial in any given context. 9 I presume that this v, merged to an R, assigns a status of "word" to the resulting R-v construct. 10 The R-v construct constitutes, to borrow Epstein et al's (2016: 98) phraseology, an "X 0 -level" amalgam "created by internal pair-Merge" (i.e. syntactic internal pair-Merge of R to v). In this 7 Chart (1) covers the major types of Jacobsen's (1992) "classification types" (except one), our Types (1A), (1B), (1C), (1D), (1E), and (1F) corresponding respectfully to his (II), (I), (III), (IX), (IV), and (VIII). The "numbers of pairs" cited without parentheses refer to the numbers given in Jacobsen's list of "core" (Jacobsen 2016: 23) intransitive -transitive pairs. The numbers in the parentheses are Matsumoto's (2016a) for the corresponding categories. For discussions on "non-core" verb pairs, see Matsumoto (2016b: 489). The one prominent type (i.e., Jacobsen's (V): e.g., kae-s 'return tr ' vs. kae-r 'return intr '; 27 pairs), and those listed in his "minor" classes are not included in (1), and will not be discussed in this paper in order to focus on transitivity suffixes AR, AS, and E. For discussions on the omitted groups, see Jacobsen (1992), Kitagawa & Fujii (1999), and Matsumoto (2016c), among others.

8
For related discussions concerning little v pertaining to Japanese transitivity suffixes, see Volpe (2005) and Marantz (2013: 106-108) among others. manner, I put into effect a perspective entertained by such works as Mendivil-Giró (2019: 1162) that "words are not lexical units, but rather syntactic constructions produced by the computational system." Japanese TrAlt is marked by the fact that the verbs that participate in TrAlt are by and large "basic" verbs (Hayatsu 1989: 250-53). 11 It is perhaps for this reason that it has inspired a great number of memorable works -including Motoori (1828), Sakuma (1936), Jacobsen (1992), Suga & Hayatsu (1995), Kageyama (1996), Miyagawa (1998), Matsumoto (2000a2000b;2016c), and to name just a few. As noted by Sakuma, Okutsu, and many others, v suffixes AR and AS show strong affinities to syntactically active direct (r)are passive and periphrastic (s) ase causative, respectively, in regard to their lexical (having to do with the surface presence or absence of the external argument) and functional (relating to [+/-acc]) properties. We begin in the next section with our investigation of AR.

AR
One critical issue about AR not yet addressed sufficiently is whether or not it is simple "intransitivizer" or "lexical passive." To pursue this issue, I will assume at the outset that Matsumoto (2000d: 175) is essentially correct in his claim that all AR's are associated with the functional feature [-acc]. The question as to whether this feature manifests as simple intransitivizer or as a lexical passive which shows notable similarity to the syntactically active direct passive (r)are (Kuroda 1979) thus depends on the distinct idiosyncratic lexico-semantic properties of the R's that select AR's.

"Lexical passive" vs. "Intransitivizer"
Recall that only 8 pairs are listed in (1) for the R-AR constructs of Type (1E) {AR/∅}. In this subsection I examine the R-AR constructs of Type (1E) {AR/∅} so as to examine the functional status of AR with respect to the feature [-acc]. 12 The set of diagnostic questions given in (2) is employed to tease out the two distinct manifestations of AR: "intransitivizer" or "lexical passive." (2) a.
Does AR, like syntactically active direct passive (r)are, promote the accusative marked NP to the subject position, demoting the agentive or triggering causer subject NP (originally case-marked nominative) to a non-subject position followed either by postposition ni 'by' or niyotte 'by'? If yes, as in (3a; 3b) and (4a; 4b), the AR functions as a "lexical passive." The structural representation of (3b) is given in (5).

b.
Does the R-AR allow the retention of the subject NP with the nominative case marking intact, demoting at the same time the accusative-marked NP to resurface as a postpositional phrase (PP)? If yes, as in (6a; 6b) and (7a; 7b), the AR functions as an intransitivizer. The structural representation of (6b) is given in (8).
Which interpretation the AR of a given Type (1E) R-AR construct would have is not predictable, being dependent on the idiosyncratic lexico-semantic property of the participating R.
In the examples to be cited hereafter, AR is simply identified in gloss as AR whereas direct passive (r) are is glossed as "passive." 11 Hayatsu (1989)  (3) Lexical passive AR: a. Kuma-ga azarasi no ko-o tukam-ta.
(AR-ta → AT-ta) seal 's baby-nom bear-by grab-AR-past 'A baby seal was caught by a bear.' (AR here is a lexical passive in accordance to (2a).) c.
Sue no kage-ga hutatu no hitokage-ni(*yotte) hasam-AT-ta. S 's shadow-nom two 's shadow-by put.in.between-AR-past 'Sue's shadow was encroached upon by two human shadows from the sides.' (AR here is a lexical passive in accordance to (2a).) c.
Sue no kage-ga hutatu no hitokage-ni(yotte) hasam-are-ta. S 's shadow-nom two 's shadow-by put.inbetween-passive-past 'Sue's shadow was encroached upon by two human shadows from the sides.'

(5)
Lexical passive AR: In (3b)/(5), a lexical passive AR with [-acc] internally pair-Merges to an R with the consequence that NP azarasi no ko 'baby seal' raises to the Spec of TP to receive the nominative case, with the agent kuma 'bear' retained in the postpositional ni 'by' phrase just as it happens with the (r) are direct passive (3c). 13 There is no licit construction where an AR occurs with R tukam 'grab' functioning as an intransitivizer.
Examples of AR behaving as an intransitivizer are given below: Dokubari-ga kare no kubi-ni sas-AT-ta. poison.needle-nom he 's neck-into stick-AR-past 'A poisoned needle stuck into his neck.' (AR here is an intransitivizer in accordance with (2b).) c.
Kare no kubi-ga dokubari-ni(yotte) sas-{*AT, (r)are}-ta. (sas-(r)are → sas-are) he 's neck-nom poison.needle-by stick-{*AR, passive}-past 'His neck was pierced by a poisoned needle.' (As marked, AR here is not a lexical passive since it does not follow (2a).) In (6) the well-formed (b) sentence shows that AR here functions as an intransitivizer. In the passive context (c), sas-AR is illicit, while passive (r)are is fine, indicating further that the AR in sas-AR functions only as an intransitivizer. The next example follows suit. Kare-ga uzukumaru Kansin no senaka-ni matag-AT-ta. he-nom crouching K 's back-onto straddle-AR-past 'He straddled the back of crouching Kansin.' (AR here is an intransitivizer in accordance to (2b).)

c.
Uzukumaru Kansin no senaka-ga kare-ni(yotte) matag-{*AT, (r)are}-ta. crouching K 's back-nom he-by stralddle-{AR, passive}-past 'The back of crouching Kansin was straddled by him.' (AR here is not a lexical passive, since it does not follow (2a).) Intransitivizer AR: A given R participating in the Type (1E) doublet pattern must therefore be registered in its lexical entry with respect to these two factors: (i) its participation in Type (1E); and (ii) its idiosyncratic choice between the two possible interpretations of AR (i.e., lexical passive or intransitivizer).

Type (1C) {AR/E} AR's are lexical passives
To determine whether or not the Type (1C) {AR/E} AR functions as a lexical passive, we use the same diagnostic tests as in (2) reproduced here as (9).
(9) a. Does AR, like syntactically active direct passive (r)are, promote the accusative marked NP to the subject position, demoting the agentive or triggering causer subject NP (originally case-marked nominative) to a non-subject position followed either by postposition ni 'by' or niyotte 'by'? If yes, the AR in question is a lexical passive.

b.
Does the R-AR allow the retention of the subject NP with the nominative case marking intact, demoting at the same time the accusative-marked NP to resurface as a PP? If yes, the AR is an intransitivizer.
We know that the AR's in (10) are lexical passives because they all meet (9a) regarding the presence of -(ni)yotte 'by' and reject what (9b) prescribes. The AR's in the pattern (10a) are exemplified in (11), allowing niyotte 'by' and rejecting the retention of the subject NP with its (original) nominative case marking intact. Note the kind of similarity and non-similarity between the AR and the direct passive (r)are.
Sono sawagi-ga {kare, *yokisi-na-katta bakuhatu}-niyotte osam-E-rare-ta. that uproar-nom he, expect-not-past explosion-by pacify-E-passive-past 'The disturbance was suppressed by {him, *the unexpected explosion}.' (The fact that (11d) with passive (r)are directly attached to R is illicit, while (11e) in which it is attached to [R-E] is felicitous, shows that in the context of Type (1C) the syntactically active passive (r)are cannot directly attach to an R).
Example (11b) with lexical passive AR is structurally represented as follows (with kare 'he' representing the case of yokisinakatta bakuhatu 'unexpected explosion' as well): The R osam 'suppress' internally pair-Merges to Type (1C) AR [-acc], the consequence of which is that object NP sawagi 'disturbance' needs to raise to the Spec of TP to receive nominative case, while the agentive kare 'he' (or, the triggering causer yokisinakatta bakuhatu 'unexpected explosion') is demoted to a non-argument position followed by postposition ni(yotte).
True to the "in-between voice" characteristic of the TrAlt system, those R-AR's that belong to the (10b) and (10c) variations essentially follow the pattern of (10a) but display varying degrees of idiosyncratic variations. Thus, consider (13b) and (14b)  The variant (10c) allows the agent in the R-AR construction to take a postpositional ni 'by', but not niyotte 'by': 14 An interesting fact about (13c) with Sue is that its acceptability increases if Sue is considered to be significantly affected by the soup. This issue, following Matsumoto's (2000c) lead, will be discussed in Section 3.3. In the same context, Sue becomes still more acceptable if suupu-niyotte 'by (means of) soup', instead of suupu-ni 'by soup', is used, emphasing the instrumental nature of 'soup.' With the R's of this group, therefore, these lexico-semantic peculiarities need somehow be registered in their lexical entries. (15) a. Ten'in-ga manbiki-o mituk-E-ta clerk-nom shop.lifting-acc notice-E-past 'A clerk spotted shoplifting.' b.
Manbiki-ga ten'in-ni(??yotte) mituk-AT-ta. shop.lifting-nom clerk-by find-AR-past 'Shoplifting was spotted by a clerk.' Besides those pairs listed in (10), there are Type (1C) (AR/E) R's like azuk 'entrust to', sazuk 'give', and iituk 'give order', which do not exactly follow the pattern of classification formulated in (9) and (10). They are all ditransitive R's involving the "Source, Goal, Theme" roles. In (16a), the Source Ken may double as the Agent, and Theme kamera 'camera' transports from the Source/Agent to the Goal Sue in some capacity.
from} camera-acc entrust-AR-past 'Sue was entrusted a camera by Ken.' (Note that the status of AR here is ambiguous: it is a lexical passive according to diagnostic test (9a), but "intransitivizer-like" according to (9b) in that Ken-{ni, kara} no longer has indirect object status.) c. *Kamera-ga Ken-ni Sue-ni azuk-AT-ta.
camera-nom K-by S-to entrust-AR-past (Intended: 'The camera was entrusted to Sue by Ken.') d.
Kamera-ga Ken-niyotte Sue-ni azuk-E-rare-ta. camera-nom K-by S-to entrust-E-passive-past 'The camera was entrusted to Sue by Ken.' The attachment of lexical passive AR to a ditransitive R like azuk 'entrust' thus strips away the indirect objecthood from the Goal argument, and promotes it to the subject position, converting the ditransitive status of this R to that of the transitive ([[R] ditr -AR] tr ). In this manner, the feature [-acc] associated with AR exposes the two faces of AR manifesting in the same context: (i) with its "lexical passive" face, it demotes the "Source" subject Ken to the postpositional Ken-{ni, kara} 'from Ken'; and (ii) with its "intransitivizer-like" face, it reduces ditransitivity to transitivity by promoting the indirect object Sue-ni to be the subject. 15 The specifics of those variations associated with the R's must be registered in their lexical entries. Here again, differences in the degree of acceptability associated with each of those cases accentuates the "in-between voice" characteristics associated with AR.

The context for "being affected"
Consider again example (13c), reproduced here as (17). To be noted here is that the sentence with the sentient subject Sue may turn out to be less odd if we assume that the mental, psychological, or physiological state of the subject referent Sue is significantly affected by the event expressed by the predicate (i.e., 'warmed by the soup').
(17) {*Sumibi, Sue}-ga suupu-ni(yotte) atatam-AT-ta. {charcoal.fire, Sue}-nom soup-by warm-AR-past '{*Charcoal fire, Sue} was warmed by soup.' (The status of AR is ambiguous: it is a lexical passive with respect to the diagnostic test (9a) but an intransitivizer according to (9b).) Kitagawa Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 In (18) too Sue is the significant "affectee" in the event of being hit by a car: (18) Sue-ga kuruma-ni(*yotte) butuk-AT-ta. S-nom car-to hit-AR-past 'Sue was hit by a car." (AR is a lexical passive with respect to the diagnostic test (9a), but is an intransitivizer according to (9b).) What is common with (17), (18b), and other similar constructions are the curious "affectee" status of the subject NP referent and the concomitant ambiguous status of AR of being either a lexical passive or an intransitivizer. This is reminiscent of the phenomenon pointed out by Matsumoto 2000c in relation to a transitive and ditransitive pair like osie-ru 'teach' and osowar-u 'learn'. Thus, consider the following pair: a.
Seito-ga sensei-ni eigo-o osowat-ta. student-nom teacher-by English-acc learn-past 'Students learned English from the teacher.' Matsumoto (2000c: 82) notes that the felicity requirement for the transitive osowar-ru and ditransitive osie-ru type pairs is that the subject NP referent of the transitive construction (which corresponds to the referent of the NP-ni in the ditransitive) must be associated with a sense of being significantly affected by the event described by the predicate. Thus, in the case of (19), seito 'student' is the significant affectee in the event "English lesson."

Productivity phenomena of AR
Transitive suffix AR engages in productivity phenomena in two markedly different manners: one associated with substandard status, and the other with no such stigma.

Type (1E) {AR/∅} AR and its substandard productivity
Recall that, of those of Jacobsen's (1992) eight pairs that belong to the Type (1E) pattern, we have identified two pairs wherein the R demands the lexical passive AR (i.e., (3b) and (4b)) and two others with which the R calls for the intransitivizer AR (i.e., (6b) and (7b)). This is the picture that fits Hasegawa's (2001: 2) characterization, namely, that Japanese TrAlt is a system wherein it is "not predictable" which suffix a particular R takes. This is the underlying assumption behind Jacobsen's (1992) "Appendix list" as well.
For Kageyama (1996: 191), these Type (1E) R-AR constructs exemplify "new word creation." With Nishio (1954), these are instances of new word creation made in "analogy" to the Type (1E) {AR/∅} pattern. How about their substandard status? The answer is clear enough. As Kageyama (1996: 191-192) points out, the reason for their substandard status has to do with the fact that the transitive R's that are involved here are those that are normally considered to "lack the intransitive counterparts." The lexical passive AR is attached illegally to those transitive R's. In these cases, the lexical passive AR masquerades as a direct passive. (Kageyama also notes that these R's constitute a natural lexico-semantic class, being of "the kind that expresses a state or a change of location.") Indeed, these R's under discussion have all the signatures of the transitive R's that "lack the intransitive counterparts" according to various empirical tests advanced in Hayatsu (1989). She observes, on the basis of such phenomena as agentive nominalization, iteration phenomenon, co-occurring adverbs, and compound verb formation, that there is a notable lexico-semantic contrast between the transitive R's that participate in TrAlt and those that do not, clustering around the aspectual sense of "result." The R's that do not participate in TrAlt tend to be those that express a sense of process; they are aspectually [-state, +durative, -telic] with the telic feature fixed as [-telic]. In contrast, the R's that do participate in TrAlt can relate to the resulting points of eventuality; it is normally the kind that is aspectually characterized as [-state, +durative, +telic]. Thus, the capacity of an R to express an ongoing process can be a test to tease apart those two types of R's (Hayatsu 1989: 240-243). And, the R's in the R-AR examples exhibiting substandard status (e.g., dak 'hug' in (20a)/(21a), nom 'drink' in (21b)) are exactly those that have the capacity of expressing an ongoing process, as the following examples show: (21) a.
Ano ko-wa sakki-kara hahaoya no syasin-o mune-ni that kid-top a.while.ago-from mother 's photo-acc chest-to daki-tuzuke.te i-ru. hold-continue.and be-pres 'That kid has been holding his mother's photo to his chest for some time now.' b.
Kare-wa hitoban.zyuu sake-o nomi-tuzuke-ta. he-top all.night.long rice.wine-acc drink-continue-past 'He just kept drinking sake all night long.' These examples demonstrate that the R's that appear in substandard R-AR's like dak-AR 'be held' are of the type that do not participate in TrAlt. Hence, dak-AR is illegal and marked as substandard. In popular usage, such substandard R-AR constructs are quite unstable. My sense is that, while an example like dak-AR 'be embraced' in (20) is fairly understandable, an example like mus-AR 'be steam-cooked' of [[mus] tr -{AR/∅}] in the following example taken from Yokomitsu Riichi's novel Hai 'Flies' (1923) (Miyaji 1985: 293) goes beyond the permissible line today: (22) ?*Manzyuu-wa mada mus-AR-an no kai? dumpling-top yet steam-pass-not is.it q (Intended: 'Are the dumplings not steam-cooked yet?') As Miyaji notes most of these substandard forms disappear in time.
What is important to note, at the same time, is that this substandard productivity phenomenon is highly symptomatic of the passive-like function of a lexical passive AR, contributing to the TrAlt's "in-between" voice status. Native speakers of Japanese are quite aware, and indeed quite ready to take advantage, of the distinct passive-like force associated with the lexical passive AR, even risking the consequence of its substandard status. That is to say, the v suffix AR here is not subjugated entirely to its host R's idiosyncratic lexico-semantic property. On one hand we do need to acknowledge that each R participating in TrAlt must be registered in the dictionary for its co-occurrence potential with AR. But, on the other, the substandard behavior of AR's noted Kitagawa Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 in this subsection shows that the same AR's can act like syntactically active passives. AR here is behaving badly, but potentially not so prohibitively. And, this, of course, is what we might expect from the Japanese TrAlt, which is an "in-between" voice system.

Productivity phenomena involving the Type (1C) {AR/E} pattern
Unlike the substandard productivity of Type (1E) {AR/∅} AR, the productivity involving Type (1C) {AR/E} AR is quite robust and stable.
Noteworthy for our purpose are the fascinating diachronic findings by Nishio (1954). He lists three historically classified periods as in (23a) (23a). This productivity process was apparently quite robust, which would attest to the fact that the type (1C) {AR/E} pattern is by far the largest group in (1). It would also lead us to speculate that there may well be far more actual cases of the Type (1C) R-{AR/E} pairs beyond those listed by Jacobsen's (1992) "Appendix." In fact, Miyaji's list of such unlisted pairs includes kagam-{AR/E} 'squat' (Miyaji 1985: 227-233), mak-{AR/E} 'reduce in price' (Miyaji 1985: 88-89), and sukum{AR/E} 'cower' (Miyaji 1985: 204-208

Lexical causative AS
In the view that has been widely accepted since Kuno (1973) and Shibatani (1973), the distinction between the "lexical causative" with AS and the syntactically active causativizaton by periphrastic (s)ase is made in terms of the monoclausal vs. biclausal characterizations gauged Kitagawa Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 by such means as reflexive zibun 'self' interpretation involving its subject-antecedent condition and the subject-oriented adverbial interpretation. The lexical causative R-AS construction (24a) is monoclausal with just one subject NP (i.e., Hanako), and the periphrastic (s)ase causative (24b) is biclausal with two distinct subject NPs (i.e., Hanako as well as Taro).
In these examples, AS is glossed simply as "AS," while the periphrastic causative (s)ase is glossed as 'cause.' (24) a.
Hanako i -wa Taro j -o zibun i/*j no heya-de nak-ASi-ta. H-top T-acc self 's room-in weep-AS-past 'Hanako i made Taro cry in her i room.' b.
H-top T-acc self 's room-in weep-cause-past 'Hanako i made Taro j cry in his i /her j (own) room.'

(26)
The R-(s)ase causative may function as "lexical causative" (R-AS) if the R represents a process whereby (a) the subject referent is affected or at least undergoes some change; (b) the event described is mostly due to "self-directed" actions; and (c) the event is the kind that is conducted with the agent herself as the recipient or beneficiary of a moved entity or influence.
If satisfied, Matsumoto notes, they exhibit syntactically "monoclausal" characteristics as judged by such tests as subject honorification, passivization of the object NP of the base R, pronominal binding, Control, and subject oriented adverbial interpretation. The fact that reflexive karezisin 'herself, himself' may refer to the subject Anpanman in (25) indicates that the sentence is monoclausal, because, as generally accepted, karezisin "must be bound within its clause" (Matsumoto 2000a: 148). So, we now have a question: what is the status of (s)ase in (25)? Is it still a syntactically active periphrastic causative?
A significant attempt to respond to such a question is Miyagawa's (1980;1984; "Paradigmatic Structure" (PDS) thesis. So, although not directly bearing upon the main issues of this paper, let me provide a brief comment. The PDS, specifying three slots (Intransitive, Transitive, and Ditransitive), purports to be a system in terms of which a verb, either simple or complex, can be "placed into an appropriate slot" (Miyagawa 1984: 186); as such, it constitutes "an organization within the lexicon that filters as to which verbs may enter the permanent lexicon" (Miyagawa 1998: 71). In (24), transitive tabe 'eat' being an R that does not participate in TrAlt, tabe-sase 'eat-cause' fits into the PDS's ditransitive slot. This system provides a principled way to explain, for instance, why the intransitive R niow 'smell' lacking the transitive partner may adopt the causative form niow-ase 'smell-cause' to constitute an idiom that means "hint" (Miyagawa 1998: 71, M's (11)).
(i) Yasya i -wa kodomo j -ni zibun i,*j -o tabe-sase-ta. Yasya-top child-dat self-acc eat-cause-past 'Yasya gave her own body to feed her children.' (Yasya is a legendary figure who fed her children with her own flesh when their lives were threatened with starvation.)

Kitagawa
Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 Miyagawa (1998) admits, however, that his PDS has difficulty in handling triplets, which involve two distinct lexical forms competing for just one PDS slot (e.g., [mor] intr 'leak intr ' and [mor-E] intr 'leak intr '; or [tok] tr 'dissolve tr ' and [tok-AS] tr 'dissolve tr '). Kageyama (1996) makes an important contribution to our understanding of the nature of causation expressed by AS. He takes advantage of the fact that some R's expressing animate motion of "self-induced change" may occur either with the AS oriented (1F) R-{∅/AS} pattern or with "transitivizing" E forming Type (1A) R-{∅/E}. Analyzing the semantic distinction between AS and E in this minimal pair context in terms of his "Lexical Semantic Relation" (LSR) framework, he concludes that AS may accommodate a "triggering causer," while E calls for a "volitive NP" as its external argument EA (Kageyama 1996: 69-77;190-198). His claim is based on examples like the following (Kageyama 1996: 197, K's (137b) and (138b) In (27a) with E, the sentient EA kodomo 'child' qualifies to be a volitive NP, while the nonsentient EA densya no sindoo 'train's vibration' does not, thus rendering the latter infelicitous. In example (27b), AS, which simply requires its EA to be a triggering causer, fits in with either animate kodomo 'child' or inanimate densya no sindoo 'train's vibration'; hence, both are felicitous in (27b). Suga (1980: 37, (58) and (60)) notes a similar contrast: Sentence (28a) calls for a volitive EA, providing a context favorable to E but not to AS. In (28b) with inanimate EA, the situation is the reverse: only AS that relates to triggering causation is felicitous. Matsumoto (2000b) presents a critique of Kageyama's claim, stating that the distinction between -e and -as is "only an accidental product of his examples," and that "the distinction comes from the semantics of the base verb rather than from the affix chosen" (Matsumoto 2000b: 178). In his view, -e and -as "both trigger causativization in a similarly indeterminate way, allowing different possibilities for the nature of their subject"; "this kind of distinction" is "a part of the regular semantics of each semantic type of verb" (Matsumoto 2000b: 181). I concur with the view that the distinction comes from the semantics of the base R. In this, however, I adopt the line of analysis suggested by Arad (1998: 124). She proposes: "we should distinguish not between agents and causers as such but between predicates selecting for an agent and predicates which allow a causer." For, "agents and causers are generated in the same syntactic position, Spec vP." Selection of v* suffix E* by V is the "something more" which distinguishes agents from causers, allowing a volitive agent in its Spec vP*. If a given R like ak 'open' selects either AS or E, the contrastive semantic characteristics associated with AS or E may then manifest as exemplified in (28a) and (28b). Following Kageyama, I thus maintain that AS may call for a "triggering causer." As for E, a detailed analysis of it follows directly. 14 Kitagawa Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 6 Multi-facet manifestation of TrAlt Suffix E As Narrog (2016: 257) points out, "a common assumption" about E from a diachronic point of view is the following: 17 the vowel -e in the verb base goes back to a morpheme *-Ci (e.g., Unger 2000: 664, 667) that Robbeets (2007Robbeets ( , 2010 labels as "causative-passive" and identifies as *-(k)i. Traditionally it has been assumed that this is related to causative-passives in Korean and other Transeurasian languages.

Two types of causation
It is this "causative-passive" suffix E, viewed synchronically, that is our topic in this section. Jacobsen (1992: 57) notes that two of the main classes of verb pairs "exhibit the same suffixes, but marking precisely the opposite sides of the intransitive vs. transitive opposition." These are the groups that are identified in (1) as Types (1A) R-{∅/E} with "transitivizing" E and (1B) R-{E/∅} with "intransitivizing" E. It is largely the difficulty of treating these E's either as one morpheme or two that has forced researchers like Okutsu (1697) to conclude that E is simply part of R. Sakuma (1936: 137) has noted: "it is clear that -ar is involved with intransitivization and -as with transitivization," but "-e is involved with both, depending on the situation; herein may lie an important clue to unlock the system of verbal constructions in Japanese." It is with this backdrop that I will attempt to present a unified analysis that these E's are in fact all instances of one and the same v suffix E.
Suffix E's chart (1) (2008) and Frellesvig & Whitman (2016) propose an alternative view that the intransitivizing E originates from "verb e-'get, obtain, be able to'." The combination of the basic root of a verb with the verb e-"results in the subregularities that survive to this day: transitive accomplishment verbs from the combination of 'get' with intransitive achievement, and a smaller number of anticausative intransitives from the combination of 'get' with transitives" (Frellessvig and Whitman 2016: 289). Whitman (2008: 165) is certainly correct when he says that "whatever morpheme was involved in the derivational relationship linking quadrigrade and bigrade stems, it was not a simple 'transitivity flipper'" (such as suggested, in my view, in an extensive diachronic study by Mabuchi 1999: 300-301 or in a synchronic account of E in Kitagawa & Fujii 1999). To be noted, however, is that one possible meaning of their "verb e-" is 'be able to'. And this is a problem. Synchronically, potential auxiliary (rar)e 'be able to' is distinct from v suffix E in one critical respect. It can attach to a transitive R that does not participate in TrAlt (e.g., nom 'drink', tabe 'eat'): (32) a. Suidookan-kara mizu-ga mor-u. water.pipe-from water-nom leak-pres 'Water leaks from the water pipe.' b.
Meibo-kara watasi no namae-ga mor-E-ru. register-from I 's name-nom escape-E-pres 'My name is not listed in the register. ' Suga (1980: 35, (35A,B)) According to Suga (1980: 35), "X-ga mor" means 'X moves out (from where it was located earlier)'. If this "movement" can be visibly traced, as in (32) with water flow, both (32a) the event of water leaking (mor) and (32b) the resultant state of water having leaked (mor-E) Kitagawa Glossa: a journal of general linguistics DOI: 10.5334/gjgl.1304 I will thus posit as a plausible working hypothesis that v suffix E is uniquely characterized by the feature [+Resultant]. In the succeeding subsections we examine how this feature [+Resultant] affects the interpretation of the R-E's depending on the given R's verb types: i.e., unaccusative, unergative, or accomplishment.

E attachment to unaccusative R's
Recall that in Section 5.1 we concurred with Kageyama (1996) that the [R-E] tr in Type (1A) {∅/E} calls for a volitive NP in the subject position. In the sense of Chomsky (2013: 43, fn. 29), the E of this type is v*. I therefore maintain that in Type (1A), wherewith E's involve transitivity directly, the resultant state is one that is effected by an outside force: (36) The feature [+Resultant] associated with E may imply an outside force effecting the resultant state.
Sue-ga asagao no turu-o ikegaki-ni karam-E-ta. S-nom morning.glory 's vine-acc hedge-to entwine-E-past 'Sue made the morning glory vine entwine around the hedge.' Unaccusative R's are inherently subcategorized with an internal argument (IA), and call for no external argument (EA). I suggest that, in the structural context of the [R-E] of such an R, v suffix E* with its [+Resulant] feature has an effect of assigning [+acc] to that IA, a consequence of which is that the IA (e.g., asagoa no turu 'morning glory vine' in (37b)) needs no longer raise to the Spec of TP position to receive the nominative case, the EA (e.g., Sue) being introduced independently as the volitional agent. (37b) is structurally represented as follows: Here, karam 'entwine' internally pair-Merges (Epstein et al. 2016: 98) to suffix E*, {v*, <karam, E*>}, to create a "X 0 -level amalgam." Aspectually, this [[ intr R]-E*] tr is [-state, +durative, +telic] (i,e., "Accomplishment"). The E attachment thus "transitivizes" an R, rendering the EA of the [[ intr R]-E*] tr to be a volitive NP.
Let me note that the discussion so far is consonant with Jacobsen's (1992Jacobsen's ( , 2016) "semantic markedness" thesis, the basic guideline of which may be stated as follows: (39) Semantic Markedness thesis: "when two related linguistic forms co-exist, one relatively more marked than the other, there will occur a corresponding division of labor in the meaning expressed by the two forms, with the marked form reserved for expressing a more "extraordinary" state of affairs than the unmarked forms" (Jacobsen 1992: 12).
Since "it is commonplace in human experience for entities possessing organic or other complex functions to lose those functions spontaneously with the passing of time, without the influence of external force," the Semantic Markedness thesis predicts that the R's that belong to Type (1A) would participate in the "causative" {∅/E} pattern (Jacobsen 2016: 28).
Sue i -ga zibun i no tumibukasa-ni kurusin-da. (kurusim-ta → kurusin-da) S-nom self 's sinfulness-dat suffer-past 'Sue suffered of her own sinfulness.' Hasegawa claims, and I concur, that the nominative subject in example (40a) "is not an external argument but is derived from inside VP" (p. 22). Supportive evidence for it is the binding phenomenon. In (40a), the antecedent Sue does not c-command the reflexive zibun 'self'. This apparent puzzle disappears if we assume that (40a) is structurally represented as in (41), in which case the antecedent Sue does c-command the trace of zibun, and we also assume that the internal pair-Merge of R kurusim 'suffer' to suffix E allows [R-E] to assign the accusative case to the IA Sue.
The fact that R-E can participate in backward reflexivization in a fashion parallel to the periphrastic causative (s)ase as in (40a) shows, once again, the "in-between voice" status (Noda 1991: 214) of the Japanese TrAlt system. 20

E attachment to unergative R's
We have seen that the [+Resultant] feature of v suffix E attached to an unaccusative R has an effect of assigning [+acc] to its internal argument (IA). The unergative R subcategorized with no IA precipitates no such operation. Aspectually, the unergative R's are [-state, +durative] with the telic feature unspecified. The E attachment to such an R renders the resulting [[R] intr -E] intr construct to be [-state, +durative, +telic], but, without the presence of IA, E fails to act as v*. Hence, the [+Resultant] feature associated with E functions only to highlight a resultative sense of the eventuality expressed by the co-occurring R (recall the discussions in Sections 6.1 and 6.2). The structural representation of (33b) (Suga 1980: 35, (34B)), repeated here as (42a), is (42b). operating between two spheres, namely, the truly idiosyncratic lexical formation expressing voice distinction AND the common syntactic operation involving direct passive (r)are and periphrastic causative (s)ase. This "in-between voice" characteristic manifests in a seemingly contradictory manner within the TrAlt system in that, on one hand, which suffix a given R takes is "not predictable" (Hasegawa 2001: 2) but, on the other, the TrAlt also allows quite robust new R-AR formations ("new" in the sense that the given R-v constructs were not registered in any relevant dictionaries earlier).
Following Nishio's (1954) diachronic study on the new R-AR creation, this paper has aimed to delineate the nature of the productivity mechanism in engendering new R-AR formations in synchronic terms. What we have found is that, synchronically, the major driving force for new word creation is not "analogy" as assumed by Nishio. "Analogy" may open a path for the derivational directionality (46) 3). In general terms, though, what functions does analogy play in the system of Japanese TrAlt? The function of analogy in this sense is to identify a possible slot (i.e., either intransitive, transitive, or ditransitive, in a manner reminiscent to PDS as envisioned by Miyagawa (1980;1984; see our Section 5.1) in the context of TrAlt that can be filled by one of the three transitivity v suffixes AR, E, and AS, in order to complete a doublet pair in accordance to the patterns represented in (1).
As to what makes the mechanisms for productivity work, there is a unique set of functions associated with two v suffixes in particular, AR and E, interacting not only with their host R's but also with each other. Suffix AR with its [-acc] feature may function as lexical passive in a manner significantly similar to direct passive (r)are, as discussed in Sections 3 and 4. Suffix E with its [+Resultant] feature interacts with its host R's lexico-semantic properties in a complex but definable manner, as discussed in detail in Section 6. Suffix E, furthermore, with its peculiar properties as articulated in (47) and (48), contributes to the proper setting of derivational directionality involved in the new R-AR creation in the context of the Type (1C) {AR/E} pattern, as we have just seen in Section 7.
The mechanisms of productivity operative in the context of Japanese TrAlt are thus built on, and sustained by, such a set of functions associated with the two suffixes, AR and E, interacting with each other, and working collectively, as a driving force in creating new word formation in a clearly definable manner.