Indonesian has a group of verbs like menganggap 'believe' that subcategorize for clausal complements. Interestingly, the complement clause of the verbs allows NP-movement out of it. To account for this peculiar syntactic behavior, the Raising to Object (RTO) analysis has been proposed. Close examination of the construction, however, provides evidence against it and gives favor to the Exceptional Case Marking (ECM) analysis.

Verbs like menganggap 'believe' in Indonesian subcategorizes for finite complement clauses as evidenced by the occurrence of modal/temporal expressions in them.

(1)	John menganggap dia (akan/bisa) membaca buku itu.
	John meN-consider 3 (will/can) meN-read book that
	'John believes that he (will/can) read the book.'

Interestingly, the embedded subject becomes the matrix subject under passivization as in (2).

(2)	Dia dianggap (oleh) John (bisa) membaca buku itu.
	3 DI-consider (by) John (can) meN-read book that
	'He was believed by John to have read the book.'

Chung (1976) and Musgrave (2001) argue that the embedded subject dia in (1) is part of the matrix clause (i.e., an RTO construction). Another supporting evidence comes from other syntactic operations such as predicate movement and cliticization where reduced pronouns (e.g.-nya 'reduced 3^rd person pronoun') are attached to the matrix object.

(3)	lalai, Johni menganggap-nya*i/j /dia*i/j /*ia
	careless, John meN-believe -3/ 3SG/ 3SG
	'John considered him to be careless.'

In (3), dia and -nya cannot refer to the matrix subject, John and ia cannot follow the matrix verb. These facts seem to show that dia is in the matrix object position.

Unlike English, however, in which NP-movement is triggered by a Case requirement, it is not obvious what triggers the rising of the embedded subject out of the finite clause. Furthermore, various constituency tests such as adverbial placement, pronominal binding, and distribution of another third person pronoun ia, show that the embedded subject is in the embedded clause and does not rise to the matrix clause object position.

(4)	Johni menganggap ia /diai/j kemarin lalai.
	John AV-consider 3 / 3 yesterday careless
	'John believed that he was careless yesterday/*John believed yesterday that he was care less.'

From (4), we see that: (i) an adverb like kemarin 'yesterday' following dia cannot be interpreted with the matrix verb, (ii) the third person pronoun, dia, can refer to the matrix subject John, (iii) another third person pronoun, ia, which occurs only in the subject position, can follow the matrix verb, indicating that dia is a subject. This array of facts, together with the lack of trigger, strongly indicates that the embedded subject remains in the embedded clause.

One might argue that further syntactic operation is a prerequisite for RTO such as passivization, predicate movement, and cliticization. However, the restriction on the distribution of reflexives and wh-phrases provides counterexamples to such an analysis. As illustrated in (5), Indonesian does not allow reflexives or wh-phrases in the subject position.

(5)	a*.	John menganggap dirinya sendiri memukul Ali.
		John meN-consider body-3 self meN-hit Ali
		'John believed himself to hit Ali.'
	b*.	John menganggap siapa memukul Ali?
		John meN-consider who meN-hit Ali
		'Who did John believe to hit Ali?'

Even in the sentences involving a syntactic operation such as predicate movement, reflexives and wh-phrases are not allowed as shown in (6):

(6)	a*. Memukul Ali, John menganggap dirinya sendiri t.
	b*. Memukul Ali, John menganggap siapa t?

Therefore, we propose that verbs like menganggap 'believe' should be analyzed as ECM verbs. In the ECM analysis, optional CP-deletion renders the matrix verb govern the embedded subject. When two potential governors compete, lexical government (i.e., the matrix verb) is preferred just like in Imbabura Quechua (Hermon 1984). Due to lexical government, the embedded subject shows the properties of both a matrix object and an embedded subject.

References

Chung, Sandra (1976). On the subject of two passives in Indonesian. In Charles Li, ed., Subject and Topic. 59-98. New York: Academic Press.
Hermon, Gabriella (1984). Syntactic Modularity. Dordrecht: Foris Publications.
Musgrave, Simon (2001). Non-subject arguments in Indonesian. Doctoral dissertation. University of Melbourne.