Austronesian languages prefer agent-first orders: Some notes on Riesberg et al. (2019)

In their 2019 paper “How universal is agent-first?”, Sonja Riesberg, Kurt Malcher and Nikolaus Himmelmann argue that western Austronesian languages provide evidence for a general “agent-first” preference, although they often have a neutral order in which the patient (or another type of “undergoer”) unexpectedly precedes the agent. They examine 51 languages of the Philippines and Indonesia/Malaysia, and they find that seven of them show a gap that may be explained by an “agent-first” force (groups 2a-2c on p. 554).

This gap concerns some Indonesian/Malaysian languages (Tondano, Kelabit and a few others), which have only three out of four orders that might be expected: [V-A]-P’, P’-[V-A], A’-[V-P], but not *[V-P]-A’. Thus, only one of the two conceivable orders with A following the P is found in these languages. Is this so significant that it is worth invoking a very general principle whose causal effect is poorly understood? (Does the effect involve topicality? Agents tend to be topics, and and topics tend to come early, even when they are not agents, so maybe it’s a topicality effect? Or is it a role-specific effect, perhaps a biocognitive constraint, as suggested by Bickel et al. (2015)?)

The paper also mentions a range of other ways in which agents tend to precede patients (and other “undergoers”). For example, in clitic pronoun sequences, there is sometimes a clear preference for “agent > patient” orders, and when the nominals carry no role-marking flags, there may be a word-order freezing effect forcing an agent-first interpretation. Especially the latter type of effect, which is widely attested across the world’s languages, is an excellent argument for a highly general agent-first preference (whatever its source), I think.

So the evidence for the agent-first pressure is good, but it has long been good: Siewierska (1988: 51-56) summarized research from the 1970s and and 1980s, and as noted, it strongly overlaps with the topic-first and the animate-first principles. Now it is true that in the western Austronesian languages, there is a special Topic function (which Riesberg et al. call “subject”), marked by ang= in Tagalog (and by similar special markers in other Philippinic languages). The Topic has no special flag in the Indonesian/Malaysian languages, but it can be recognized by its position – not necessarily adjacent to the verb and its associate. (In the second paragraph above, the verb-associate combination is marked by brackets, and the Topic by a prime.) One can perhaps say that this special feature of these languages (associated with “symmetrical voice”) gives us privileged information about the agent-first preference, independently of a topic-first preference (though not of an animate-first preference). However, recall that an asymmetry in neutral orders of full nominals is seen only in seven out of the 51 languages.

It thus seems to me that perhaps the most interesting aspect of the paper is a diachronic story about the Philippinic languages that Riesberg et al. hint at in §2.4. Apparently, Tagalog represents the original situation (perhaps close to Proto-Malayo-Polynesian?): The only common orders are V-P-A’ for the Actor voice, and V-A-P’ for the Patient voice (and similarly for other undergoer voices). These two are exemplified in (5a-b) (the flag ang= is the Nominative marker, indicating the Topic role).

The Actor voice thus only allows the V-P-A’ order, which is dispreferred by the agent-first force. And some related languages have changed this, allowing V-A’-P as well in the Actor voice (e.g. Ibaloi), or even requiring V-A-P in both voices (e.g. Tatana’). So if the Tagalog situation is the original one, here we have a change from a dispreferred situation to a more preferred one. So far so good.

But how did the dispreferred situation come about? Riesberg et al. say that there is a language-particular “VP” constituent (p. 548), binding together the verb with its non-Topic argument, as illustrated by the notation “[V-A]-P’” above. However, this cannot be a competing causal force (neutralizing the agent-first principle in some cases), because it is not cross-linguistically general in any way. In fact, very few languages have any evidence for a “VP” constituent containing the verb and its agent (whereas evidence for verb-patient “bonding” is of course quite strong, as described by Tomlin’s (1986: 77-101) verb-object bonding principle). And Riesberg et al. themselves note that there is no clear evidence for a “VP” constituent in Tagalog. So why is Tagalog so strict about its V-P-A’ order in the Actor voice (in 5a above)?

The reason could be that all Tagalog (and Philippinic) main-clause constructions originate in a nominalization construction (cf. Kaufman 2009): Clauses like ‘Teacher reads paper’ come from something like “Reader-of-paper (is) teacher” (involving an agent nominalization), and clauses like ‘Cat ate rat’ come from something like “Eaten-thing-of-cat (is) rat” (involving a patient nominalization). If the predicate comes first, we get exactly the order that is the most common order in Tagalog (with the Topic in final position). The other Philippinic languages (such as Ibaloi and Tatana’) would then gradually have introduced more normal orders, more respectful of the agent-first principle. This scenario would not require any competing causal force. (The Indonesian/Malaysian languages seem to have introduced Topic-first order in addition to the original Topic-last order, cf. p. 536, perhaps because of the general topic-first pressure; but why was this not in force all along?)

In the 19th and much of the 20th century, a diachronic scenario of this kind would have been of great interest in itself. But in the current climate (where the prestige of general linguistics reigns supreme), it seems that one needs to make synchronic universal claims. This does not work very well if one’s domain is a set of 51 rather closely related languages, but Riesberg et al. nevertheless try to formulate a synchronic universal generalization (p. 526):

(4) Universal agent-first hypothesis (for symmetrical-voice languages)

In symmetrical-voice languages, deviations from the basic word-order set in transitive constructions will be biased in favour of agent-first orders such that divergence from the basic set resulting in agent-first is possible, while divergence leading to undergoer-first is disallowed.

I tried to understand this but failed. What is the “basic word-order set”, and what is meant by “deviations/divergence”? Riesberg et al. explicitly exclude non-neutral orders (with question-word fronting, clefting, etc.) from consideration, so it seems that (4) can be made sense of only if interpreted diachronically, as sketched above.

Earlier in the paper, they use “basic word order” in the usual sense (roughly the same sense as dominant word order), as the most common word order to which alternatives may occur, but less frequently:

“For an areally and genetically [sic] well-balanced sample of languages showing one of these basic orders, the hypothesis would predict that variants allowing agent > undergoer orders as alternatives to a basic undergoer > agent order are considerably more common than variants allowing undergoer > agent orders when the basic order is agent > undergoer.” (Riesberg et al., p. 526)

But for their Austronesian languages, they explicitly say that they include both the Actor voice and the Patient voice and regard both as basic. So what exactly is a “basic order” (note that they also talk about “default”, “standard” and “unmarked” orders, without explaining these notions; they do not use my term “neutral order”). And what is a “deviation” or “divergence”? It appears from several passages in §2 that they actually mean a diachronic change away from the Tagalog situation, which is thought to be the original one. (E.g. §2.3: “Another deviation from the basic verb-initial + final subject word-order set shown in (6) that results in a generalized agent-first word order involves the obligatory occurrence of postverbal agent clitics in the actor voice [in Tboli]…”)

Thus, the authors are not fully explicit about their comparative concepts, perhaps because their enterprise is really a diachronic one, even though they do not say so very clearly. The crucial notions of “Actor voice/Patient voice” and “Topic” (= their “subject”) are not well-defined as comparative concepts (=uniformly identifiable in all languages) either. In this case, this does not matter so much, because the cognacy of these notions across Philippinic and Indonesian/Malaysian languages has long been evident. When cognacy is evident, we can use proper names for the cognates (as with words, e.g. “the rọka ‘hand’ word”, for the Slavic cognates ręka (Polish), ruka (Russian), etc.) and do not need uniformly defined comparative concepts.

In summary, this is an interesting paper about some diachronic trends in western Austronesian (as well as a range of other phenomena that have close parallels around the world), even though it does not really take us closer to understanding the nature of the agent-first principle. But progress in comparative linguistics has rarely been fast, and it is reassuring that so much of what we find here is in line with what comparative linguists have long been saying (and as is summarized in Tomlin’s 1986 and Siewierska’s 1988 monographs). Consolidation of earlier positions is an important goal of research, and since the authors cite neither Tomlin nor Siewierska, their findings may even have been reached somewhat independently (thus providing even more convincing support).


Bickel, Balthasar & Witzlack-Makarevich, Alena & Choudhary, Kamal K. & Schlesewsky, Matthias & Bornkessel-Schlesewsky, Ina. 2015. The neurophysiology of language processing shapes the evolution of grammar: Evidence from case marking. PLOS ONE. Public Library of Science 10(8). e0132819. (doi:10.1371/journal.pone.0132819)

Kaufman, Daniel. 2009. Austronesian Nominalism and its consequences: A Tagalog case study. Theoretical Linguistics 35(1). 1–49.

Riesberg, Sonja & Malcher, Kurt & Himmelmann, Nikolaus P. 2019. How universal is agent-first? Evidence from symmetrical voice languages. Language 95(3). 523–561. (doi:10.1353/lan.2019.0055)

Siewierska, Anna. 1988. Word order rules. London: Croom Helm.

Tomlin, Russell S. 1986. Basic word order: Functional principles. London: Croom Helm.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.