Note: I wrote this reflection for a class some time ago. Now I realized that the syntactic analysis in the following article is not correct.
Gibson, E., Piantadosi, S.T., Brink, K., Bergen, L., Lim, E., & Saxe, R. (2013). A noisy-channel account of crosslinguistic word-order variation. Psychological Science, 24, 1079-1088.
Gilbson et. al. (2013) reminds me of Master Yoda from the Star Wars universe, who has a preference for OSV structures. “Patience you must have.” “Your father he is, defeat him you must.” These are a few of Yoda’s masterpieces. Yoda’s speech sounds alien because the OSV structure is rarely used as a canonical word order in human languages. According to Tomlin (2014), less than 1% of the known human languages use OSV canonically. I emphasize “canonical” here because un-canonical usages of OSV are possible in many languages. The un-canonical usage of OSV in English always involves the intentional focus of the “O”. For example, I can emphasize the number of votes Trump received by saying “100 million votes, Trump got!” But this utterance is un-canonical and the object “100 million votes” has to be accented. From the perspective of the noisy-channel theory, OSV is equally inefficient as SOV, because the absence of either the noun phrase for O or the noun phrase for S will result in confusions about whether the noun phrase got left is a patient or an agent.
The problem at this point is why Gibson et. al. (2013)’s participants prefer SOV over OSV, given that both of the two structures are equally inefficient. Or, in the case of O being inanimate, OSV seems to be equally efficient as SOV. It is really puzzling as to why English-speaking participants in Gibson et. al. (2013) used SOV but not OSV. If Akhtar (2001)’s data-driven model is correct, then we would expect English-speaking adults to use OSV instead of SOV, because OSV, however un-canonical it is, actually exists in English, while SOV never occurs in English. Gibson et. al. (2013) stated at the beginning of the paper that SOV is somewhat a default word order in human languages, which is another way to say that SOV is part of the universal grammar. However, the notion of SOV being a default word order is probably against the general implication of the noisy-channel theory. As Shannon’s theory of communication requires efficiency, we would expect a default word order to be efficient if Shannon’s theory indeed applies in human communication. However, the SOV structure is apparently less efficient than the SVO structure for the reasons given on p.1080. Therefore, I think there are two problems with the data and arguments presented in Gibson et. al. (2013). (1) Since SOV and OSV are equally efficient (or inefficient), why do people prefer the former over the latter? (2) Since SOV is less efficient than SVO, why do people use SOV as the default?
The first question cannot be answered by either the noisy-channel theory or the data-driven model. The second question might be answered by taking into account the animacy of Os. Gibson et. al. (2013) show that SOV is preferred when the O is inanimate; while SVO is preferred when both S and O are animate. The claim is that an inanimate O in SOV will not cause confusion even if the S is missing, because inanimate objects cannot function as an agent. This argument is not necessarily accurate. Considering the following two sentences:
(a) The ball hit[agent] the goalkeeper[patient]. (b) The ball[agent] the goalkeeper[patient] hit.
“The ball” is inanimate, and yet it functions as agent in (a). (b) is the SOV version of (a). If either “the ball” or “the goalkeeper” is missing from (b), the intended meaning will be very hard to recover. However, if SVO is the default word order (i.e. example (a)), then the absence of “the ball” or “the goalkeeper” will not confuse people that much, because SVO tells us that whatever occurs on the left side of the verb is the S and whatever occurs on the right side of the verb is the O. In other words, SVO is much efficient than SOV even when O is inanimate. Choosing SOV as the default is in fact choosing inefficiency over efficiency, which is against the efficiency requirement of the noisy-channel theory.
Although the noisy-channel theory and the data-driven model cannot sufficiently account for the aforementioned two problems, the parameter setting theory is fully equipped to deal with these questions. I am adopting a simplified X-bar theory to show the solutions to problem 1 and problem 2. In the following syntactic trees, (3) shows the structure of SVO sentences, (4) shows the structure of SOV structures. V’ is the syntactic head of a verb phrase (VP).
(3) SVO structure
(4) SOV structure
The parameter setting of VPs requires that S occurs as a specifier of VP, which means S cannot occur in between V and O. In other words, OSV structure is highly problematic if not impossible. Yoda’s syntax is thus very alien. For SVO structures, the locations of syntactic heads are inconsistent. The head of VP is on the right, but the head of V’ is on the left. For SOV structures, the locations of syntactic heads are consistent. VP has its head on the right, and V’ has its head on the right too. Since (4) is syntactically more consistent than (3), the preference for (4) over (3) in fact reflects people’s bias for regularity. In conclusion, I appreciate the statistical learning paradigm and the application of information theory in language analysis, but the value of the parameter setting theory should not be categorically disregarded.
Tomlin, R. S. (2014). Basic Word Order (Reprint edition). Routledge.