TRANSFER OF L1 ARGUMENT STRUCTURE AND MORPHOLOGY IN L2 JAPANESE
This paper reports the results of an experimental study which investigates the effect of L1
Argument Structure and morphology on the acquisition of the transitivity alternation (e.g.
John broke the dish vs. The dish broke) in L2 Japanese by L1 speakers of English. English
and Japanese differ in the types of verb which admit the transitivity alternation and the way
the alternation is morphologically marked. Major findings are as follows: (1) the emerging
L2 argument structure did not override the transferred L1 argument structure, which leaves
even advanced learners in the state of optionality in L2 argument structure; (2) however, the
results of the psych-verbs which show a distinctive developmental pattern implies an
involvement of the UG property, i.e. the thematic hierarchy, as well as the L1 effect of a
The transitivity alternation in Japanese has been widely acknowledged as the most
challenging area of grammar to L2 learners (e.g. Ishikawa 1991, Mizutani 1994 among
others), mainly because Japanese transitivity alternation is mediated by myriad-forms of
morphological affixation, thus its difficulty has been assumed to be solely attributable to the
complexity and irregularity of the derivational morphology involved. However, there is
another factor involved; verbs in a given language do not always exhibit the same argument
structure pattern as their equivalents in other languages, and the transitivity alternation
demonstrates the prominent manifestation of this cross-linguistic variation between English
This paper therefore examines how this difference affects the L2 argument structure of
L1 English learners. Section 2 introduces the cross-linguistic variation between Japanese
and English in terms of transitivity alternation. An overview of previous studies dealing
with the L1 transfer of argument structure and overt morphological marker which marks
argument structure change is given in Section 3, and Section 4 presents the hypotheses of this
study. The experimental method and its results are given in Sections 5 and 6 respectively,
while Section 7 discusses the results in the light of the specific hypotheses, followed by the
2. The transitivity alternation - Japanese and English
Transitivity, a linguistic representation of whether a change in the real world is brought
about with or without influence of an external force, is a universal interpretable feature
marked on the verbal predicate. Cross-linguistic study on transitivity alternation which
suggests that the alternation typically is mediated by overt morphology in most languages
(Haspelmath 1993) offers morphological evidence of the underlying syntactic head which is
responsible for the transitivity (e.g. Hale & Keyser 1993, Harley 1995). This section
introduces the cross-linguistic variations between English and Japanese regarding the
transitivity alternation from two perspectives: how each language expresses the transitivity of
their verbs, and the distribution of transitivity alternation among verb types.
Whilst English has no overt morphological derivation process associated with
transitivity alternation, which is typologically quite unique, it is commonly observed in many
languages (e.g. French, Russian, German, etc.) that the alternation pair is morphologically
related, although the direction of derivation is subject to cross- and often intra-linguistic
variations. Japanese is one of the languages that have rich overt morphology to mark
transitivity. Transitive (Vt) and intransitive (Vi) sentences in Japanese and English are
exemplified in (1) and (2) respectively.
John-ga doa-o kowa-shi-ta
(2) doa-ga kowa-re-ta
John-NOM door-ACC break-TRN-past
door- NOM break-INT-past
‘John broke the door.’
‘The door broke.’
It can be seen that Japanese employs morphological markings on the verbal root √KOWA-
‘break’ in both Vt (-shi-) and Vi (-re-), as well as the overt case markers. The derivational
morphologies which mark the verb’s transitivity show fairly complex phonological variations
and thus it is not predictable which suffix a particular verb root will take. Furthermore, the
derivation often occurs in both directions, i.e. one of the pair is not necessarily derived from
the other. The generalization of the patterns is possible in terms of transitivity, i.e. which
form represents Vt and which represents Vi: ‘every suffix involved in transitive vs.
intransitive opposition containing [s] is transitive, and affixes containing [r] are
preponderantly intransitive’ (Jacobsen 1992: 59). From a semantic point of view, what each
morpheme adds to the root is strictly either BECOME or CAUSE. In this sense, they are
purely functional. In short, Japanese has overt morphologies to mark transitivity alternation
which do not, however, exhibit any single default pattern. In this sense, the morphological
difference between English and Japanese can simply be generalized as [± morph].
In Japanese linguistic literature, due to its non-productive nature, the transitivity
alternation has been assumed to be a lexical operation (e.g. Kuroda 1965). In this view, both
a root and Vt/Vi derivational morphology appear as a single lexical V head. However,
Miyagawa’s (1994, 1995) studies provide evidence that the Japanese lexical causative (i.e.
transitive verb) is syntactically derived. In this study, following Harley’s (1995) analysis in
the framework of Distributed Morphology (Halle & Marantz 1993), I assume Japanese Vt/Vi
derivational morphology to be an overt reflex of a functional event head, that is, the
transitivity alternation is a syntactic operation. Distributed Morphology claims that the
lexicon of a particular language consists of roots and functional features; lexical categories,
such as ‘noun’ and ‘verb’, are purely derivative in syntax. Event semantics (e.g. causation)
is determined by little v(s) rather than by a root, therefore the argument structure of a given
verb is derived from the syntactic structure into which the root is inserted. Although English
lacks morphosyntactic evidence of such a functional head, Japanese Vt/Vi derivational
morphology is arguably the overt realization of features of the little v head (Harley 1995,
In this framework, event introduction is realized by a single vP alone for the
transitivity alternation verbs. In the case of the Vi, vP is headed by BECOME v and the
complement √P, which essentially denotes the end result of the change of state introduced by
the v head. In the Vt counterpart, head of vP is replaced by CAUSE v which selects the
external argument in its Spec. Examples (3) and (4) illustrate the vP structure of the
transitivity alternation verb ‘break’, from the examples (1) and (2), in English and Japanese.
It is certainly possible to introduce additional event by projecting vP2 on top of vP1 (i.e.
syntactic causative or passive). Such structure is, however, necessary interpreted as indirect
event since the structure involve two event heads. I refer to a head of vP1 as ‘root-taking’
morpheme and a head of vP2 as ‘category-taking’ morpheme (Marantz 2001). Unlike
‘root-taking’ morpheme, which can be semi-productive due to an idiosyncratic feature of the
root, ‘category-taking’ morpheme shows complete productivity since the feature it sees
locally is functional drawn from the universal syntactic feature set (e.g. CAUSE, BECOME
(3) English (4) Japanese
(John) v CAUSE / BECOME
(John) v CAUSE / BECOME
doa √kowa- Japanese is a head final language
2.2. The distribution of transitivity alternation
The Japanese transitivity alternation is less restricted than English due to the existence
of rich overt morphology (e.g. Jacobsen 1992). The Japanese transitivity alternation applies
not only to prototypical unaccusative verb (e.g. so called change of state verb such as break,
open, melt) but also transitive verbs which denote ‘change’ (e.g. cut, decide) as well as
non-alternating unaccusatives verbs (e.g. fall, escape, disappear). On the other hand,
English transitivity alternation is more permissive allowing psych-verbs (e.g. surprise) to
alternate which do not alternate in Japanese. Although a few function as Vi by themselves
(e.g. worry), their Vi counterparts are generally available via a morphosyntactic modification
known as ‘get passive’ (5b). Japanese psych-verbs are, on the other hand, Vi version only
which do not require any additional morphology (6).
(5) a. The lion frightened the hunter
b. The hunter got frightened.
John- NOM surprise-past
‘John got/was surprised.’
Following Haegeman’s (1985) syntactic analysis, which argues that ‘get’ is not a passive but
rather that it has an unaccusative meaning1, Montrul (2001a) assumes that ‘get’ can thus be
considered as an ‘anticausative’ marker. Building on this, I assume that ‘get’ is an overt
realization of a functional BECOME v head, i.e., ‘get passive’ is a single vP. Table 1
summarises the distribution of the transitivity alternation in both languages. It is noticed that
these cross-linguistic variations provide an interesting situation regarding learnability: English
learners of Japanese are going from a subset language to a superset language with verb type 2
and 3, but from a superset language to a subset language with verb type 1.
Table 1: Distribution of transitivity alternation
Type 1: Psych-verb (e.g. surprise)
Type 2 (e.g. cut)
Type 3 (e.g. disappear)
* Overt morphology
3. Previous work on L1 morphology / argument structure transfer
Montrul (2000, 2001a) reports that a morphological pattern which marks the
transitivity alternation that exists in the L1 will transfer to the L2. Montrul conducted three
1 Haegeman proposes that ‘get passive’ (e.g. They got arrested) is the accusative version of ‘get causative’ (e.g.
He got them arrested). The difference is due to the involvement of the different type of get, i.e. unaccusative
get or causative get.
sets of bi-directional experimental studies involving English, Spanish and Turkish, and found
that L2 learners generally chose a morphological pattern, which, although possibly illicit in
the L2, which corresponded to their equivalent L1 equivalent pattern. For instance, Spanish
requires the overt anticausative marker se (pattern (b) in Table 1) for psych and COS-verbs.
Montrul found that L1 English L2 Spanish learners were more accurate in accepting Vi
version of psych-verbs with se (El cazador se asustó: The hunter got frightened) than in Vi
version of COS-verbs (La ventana se rompió: The window broke). She argues that this
divergence can be accounted by the L1 transfer of overt morphology: English requires ‘get’
with most Vi version of psych-verbs but not with COS-verbs. Whong-Barr & Schwartz’s
(2002) study on child L2 acquisition also reports that L1 Korean overt morphology affects L2
English argument structure. Korean is more permissive than English regarding the
for-dative alternation, if the benefactive verbal morpheme cwe- is added. In such a situation,
it is possible that the L1 Korean children incorrectly accepted illicit patterns in the L2, due to
L1 influence; however, this was not the case. They argue that the absence of L2 overt
morphology which marks the alternation blocks the transfer of L1 argument structure.
It has been reported that L1 influence persists when the L1 permits a superset
argument structure of the L2 counterparts (e.g. White 1991, Juffs 1996, Izumi & Lakshmanan
1998). This would occur because L2 input confirms that constructions possible in the L1 are
also possible in the L2, but never tells them that some constructions in the L1 are impossible
in the L2. Since L2 data always match L1 patterns, learners may wrongly assume that L1
and L2 are identical and overgeneralize their L1 argument structure. If language acquisition
proceeds on the basis of positive evidence only, this is a problematic case since learners must
become aware of the gap and recover from the overgeneralization without negative evidence.2
For instance, English grammar is wider than French since double object dative form is
permitted only in English (a: John sent Mary some flowers. vs. b: *Jean a envoyé Marie des
fleurs). White (1991) reports that L1 English L2 French learners who had experience of
long term immersion classes still accepted sentences such as sentence (b) above, in French.
Similarly, Inagaki (2001) reports that advanced L1 English L2 Japanese learners still accept
Japanese equivalent sentences such as ‘John walked to school’, which is illicit in the L2,
because Japanese does not allow goal prepositional phrases appearing with manner motion
verbs (e.g. walk, run). What both studies suggest is that, as predicted, L2 learners continue
to overgeneralize an L1 argument structure which is ungrammatical or illicit in the L2, even
after long exposure to the L2 input in L1 superset-L2 subset conditions.
2 In order to overcome this problem under ‘positive evidence only’ assumption, the Subset Principle - a learning
principle which requires that learners assume the most restrictive grammar until the direct positive evidence is
heard - has been proposed (Waxler & Manzini 1987). However as far as argument structure is concerned,
learners do not obey this principle, since overgeneralization is well attested in both L1 and L2 acquisition
literatures (see L1 overgeneralization in Bowerman 1974, Load 1979 among others).
In contrast, in the opposite condition, where the L2 constitutes a superset argument
structure of its L1 counterpart, the learning is expected to be less problematic, because
positive input, which indicates the existence of the L2 superset argument structure that is not
present in the L1, is available. Yet there are also cases of undergeneralization; i.e. L2
learners who failed to widen their L1 argument structure despite the existence of L2 positive
input (Montrul 2001b, Inagaki 2002). This suggests that the mere existence of positive input
in the L2 may not guarantee the broadening of the L1 argument structure; positive evidence
must be simultaneously frequent and clear (e.g. with morphological evidence).
4. Learning tasks and research hypotheses
The present study tests two issues in L2 Japanese: the effect of L1 argument structure,
and the effect of L1 morphology. In transitivity alternation, Japanese and English argument
structure exhibit two situations; (1) L2 superset-L1 subset; and (2) L1 superset-L2 subset
(psych-verbs) (see Table 1). Given that Japanese transitivity alternation is fairly prevalent in
the input and there are virtually no Vt/Vi pairs which are phonologically identical, learners
will notice the positive evidence (e.g. existence of Vi of ‘cut’ or Vt of ‘disappear’). When
learners notice such input, they are faced with two tasks: (1) they have to acquire certain type
of COS-verbs, which in English are specified for either BECOME (type 3) or CAUSE (type
2), can be underspecified for both heads in Japanese, i.e., they have to widen the L1 argument
structure, and (2) they have to learn each idiosyncratic morphological reflex of a v head which
is selected by a specific root.3 Psych-verbs, however, require different types of tasks. Since
English has wider argument structure here, learners must recognize two facts; (1) an L2
semantic equivalent of an L1 psych-verb is not compatible with a root-taking CAUSE v head,
i.e., they must narrow the L1 argument structure; and (2), the Vi counterpart does not have an
overt reflex for root-taking BECOME v head. Positive evidence will be available for (2) but
not for (1). In the light of the studies reviewed, the following two hypotheses are hereby
presented in this study.
Hypothesis 1: there will be evidence of transfer of the L1 morphological pattern.
Hypothesis 2: there will be evidence of an initial L1 argument structure transfer but the
learner will acquire the L2 specific argument structure due to positive input.
5. The study
A total of 42 learners of Japanese of three different proficiency groups, and a control
3 Whether learners can acquire individual reflex of little v is set aside as a mapping problem between syntactic
features and morphology and not directly counted as a variable in this paper.
group of 12 native Japanese speakers participated in the study. Most of the participants were
undergraduate students studying Japanese as a second language in various UK universities,
and English-speaking Japanese language tutors who teach Japanese at different educational
institutions in the UK. All participants started learning Japanese after puberty in a classroom
setting and 86% (36/42) of them had experience of living in Japan. They were first given a
proficiency test which was extracted from an earlier version of the ‘Japanese Language
Proficiency Test’ (JLPT)4, and were then divided into three different proficiency groups
according to their scores (Low-intermediate=12, High-intermediate=16 and Advanced=14).
5.2. Material and procedure
The data was gathered by an acceptability picture judgement task via the internet.
Test items consisted of the three types of verbs, shown in Table 1, with 5 tokens for each type,
whose English equivalents are given in (7).
(7) Tested items :original items were in Japanese
disappoint, surprise, please, frighten, annoy
L1 Vt/L2 alternatings cut, find, deliver, catch, dirty
L1 Vi/L2 alternatings occur, disappear, escape, fall down, fall over
In order to manipulate the argument structure of these verbs, they were embedded into four
different constructions: Vi, passive, Vt and syntactic causative, which contrasted the
‘root-taking’ vs. ‘category-taking’ morpheme choice in each Vt (CAUSE) and Vi (BECOME)
(8) Test Design
Vi-[root taking BECOME vs. category taking BECOME (-rare: passive marker)]
Vt-[root taking CAUSE vs. category taking CAUSE (-sase: syntactic causative
(9) Examples (‘disappear’): original sentences were in Japanese
The rabbit disappeared. vs. The rabbit was disappeared.
The magician disappeared the rabbit. vs. The magician made the rabbit disappear.
A picture, which depicted either Vt (with agent) or Vi event (without agent), provided the
context; this was followed by a pair of sentences: under the Vi pictures, an intransitive and a
passive sentence are provided (e.g. 9a), and under the Vt pictures, a transitive and a syntactic
causative sentence are provided (e.g. 9b). Each verb appeared twice in both of the Vi (8a)
4 JLPT is a standardized test which has four levels of proficiency: from beginner (level 4) to advanced (level 1).
I selected questions from level 3 (low intermediate) and level 2 (high intermediate).
and Vt (8b) configurations as in (8); the learners were asked to judge the acceptability of the
pair of sentences presented in the context of the pictures by using a 7-point Likert scale
ranging from -3 (completely unacceptable) through 0 (neutral) to +3 (completely acceptable).
The idea of this manipulation is that where English lacks either Vi or Vt verbs, it makes use of
a syntactic causative or passive to fill such a gap. If learners show a clear preference, for
instance in the Vt context, for the syntactic causative version of ‘disappear’ over Vt
‘disappear’, this could be interpreted as L1 influence, because ‘disappear’ in English is a
non-alternating unaccusative root, i.e. specified for root-taking BECOME, thus requiring a
category-taking CAUSE to add the causation.
In total, 30 pictures with 60 sentences were tested. All items were randomized and
there were two versions of the test. Since the target sentences included passive and syntactic
causative constructions, it was also necessary to ensure that participants had general
knowledge of passive and syntactic causatives in Japanese. Therefore, 14 distracters served
as a syntactic test. A passive and a syntactic causative sentence were paired for the distracter
items in which only one of them was unambiguously appropriate to the picture. Participants
who scored over 60% on the distracters were retained for the analysis.
As the error bar charts, given in Figure 1, show, each verb was embedded into four
different constructions (intransitive, passive, transitive and causative), and the responses for
each construction are distributed on the X-axis. The Y-axis represents the mean acceptance
rate according to the Likert-scale (from -3 to +3). Each bar represents one proficiency group
of L2 Japanese and each dotted-line box represents the appropriate judgments in L1 English.
Contrary to hypothesis 1, learners accepted the psych-verbs without additional
morphology in Vi. A group effect was detected (χ2=15.171, df =3, p=.002); the low
intermediate group was significantly less accurate than the control group (U=20.000, p=.002)
with no differences the learners, although even the low intermediate group’s acceptance rate
of the sentence was surprisingly high; 82% (of which 64% was rated as +3). In contrast to
the Vi construction, which did not confirm the expected L1 transfer in argument structure, all
groups failed to reject the inappropriate passive sentences (χ2=21.662, df =3, p=<.0001).
All learners were significantly worse than the control (p<.0001) with no differences among
the groups. The acceptance rate of the illicit passive, including the neutral value, was
consistently high among the groups; 66% for the low intermediate, 52% for the high
intermediate and 53% for the advanced group. Paired-samples t-tests performed on each Vi
and passive pair, revealed that, except for the low intermediate group (p=.073), learners
discriminated between the two constructions. i.e., licit vs. illicit sentences (p<=.0001).5
Figure 1: Mean responses on psych-verb (e.g. surprise, disappoint)
Vi Passive Vt causative
Turning to the Vt construction, a group effect was once again detected (χ2=16.793, df
=3, p=.001). Both intermediate groups were significantly worse in rejecting the
ungrammatical Vt sentences than the control group (U=18.000, p=.001; U=36.000, p=.004
respectively). This indicates a manifestation of the expected L1 transfer in argument
structure, although it did not persist in the case of passives, since the advanced group
performed in a similar manner to the controls. Finally, the results for the syntactic causative
also showed a significant difference among the groups (χ2=18.046, df =3, p=<.0001).
Similar to the Vt construction, the responses from the two intermediate groups follow the
same pattern, (as such forming a homogenous group), this is in contrast with the responses
from the advanced group which pattern with those of the control group; they were less
accurate than the controls (U=24.000, p=.005; U=24.000, p<=.0001 respectively), with no
difference among the learners. A paired-samples t-test confirmed that all of the groups
distinguished Vt vs. syntactic causative sentences (p<=.0001).
6.2. L1-Vt/L2-alternating vs. L1-Vi/L2-alternating
In general, and consistent with the predictions, learners showed a problem with the Vi
configurations (Vi vs. passive) but not with the Vt configurations (Vt vs. syntactic causative)
with type 2 verbs. The situation was reversed with type 3 verbs (see Table 1).
5 Although it is possible that these comparisons would not have been significant if the learners had not
performed so well on the Vi constructions.
Type 2 verbs (e.g. cut, find)
The results for the Vi construction shows a group effect (χ2=16.566, df =3, p=.001), in
which both intermediate groups are significantly worse than the controls (U=24.000, p=.005;
U=24.000, p<=.0001 respectively). Regarding the passive construction, the native speaker
judgments show somewhere between acceptance and rejection. Generally, the passive use of
the alternating verbs is not totally inappropriate for the description of the Vi pictures, as long
as an external causer is conceptualized (this could be subject to individual variations) in the
event; nonetheless the preferred construction for the Vi pictures was unquestionably Vi, since
the standard deviation for the Vi construction was zero for the native speakers. As Figure 2
indicates, all L2 groups in fact accepted both the Vi and the passive sentences as the
descriptions of the Vi pictures. Paired-samples t-tests for each group confirmed this; the
responses to the sentence pairs were not significant for any of the L2 groups, suggesting that
learners failed to distinguish between the two constructions regardless of proficiency level.
No group effect was found in the Vt construction. Unexpectedly, however, a group effect
was detected in the syntactic causative (χ2=112.754, df =3, p=.005), due to the low
intermediate group’s performance (U=15.000, p<=.0001), who accepted the illicit syntactic
Figure 2: Mean responses on L1-Vt/L2-Alt.
Figure 3: Mean responses on L1-Vi/L2-Alt.
Vi Passive Vt causative
Vi Passive Vt causative
ype 2: e.g.
(Type 3: e.g. occur, disappear)
Type 3 verbs (e.g. occur, disappear)
Since this verb type is a non-alternating unaccusative in English, a difference was not
expected in the Vi configuration; a prediction which was confirmed. No group difference
was found for either the Vi or the passive constructions: all learners accurately accepted the
Vi and rejected the passive. In contrast, the expected difficulty was attested in the Vt
configuration. A strong group effect was detected for the Vt construction (χ2=26.241, df =3,