There are two dominant theories that try to explain why children are able to learn their mother tongue with such little apparent effort. One proposes that a system for using and learning language is built into the brain and consequently produces some linguistic universals. The other says that language and brain co-evolved, so that the two are adapted to each other. Instead of learning inborn universals, children learn languages quickly because they evolved to be learnable. Finding the correct solution to the ease-of-learning puzzle has strong implications for how language began.
The basic difficulty of the inbuilt-system theory is the enormous variety of ways to organize language. A child learning French has a whole system of gender nouns to learn while a Zulu child has a completely different system of noun classes to master. An English child learns to use prepositions while a Japanese toddler learns postpositions. The number of such variations is quite large, much larger than the number of universals. One solution proposed has been to lump variations together. Thus, if a language has feature X it will also have features R, S, T, U, V and Y. If a child is confronted with a language that has feature X, it's brain forms a circuit that automatically brings RSTUVY along for the ride.
This idea has born some fruit. The difference between pre- and postpositions, for example, lies in where the word goes in a phrase. Prepositions go at the front ("through his heart") while postpositions go at the end ('his heart through"). English, which uses prepositions, also puts the verb between subject and object ("I shot an arrow through his heart"), while Japanese, which uses postpositions, puts the object before the verb ("I an arrow shot his heart through"). So, instead of having to learn the subject-object-verb order and use of postpositions separately, a Japanese child might notice the SOV word order and bring postposition into the system automatically.
Frankly, to me the system of features and parameters has always seemed a straw desperately grabbed. Setting parameters is common enough in computer software that offers many options, but why would a system evolve it? Why not just hardwire one solution? But my instincts are not evidence and the parameters system has many supporters. Now, however, a recent letter in Nature offers a stronger argument against the parameters theory. The authors (Michael Dunn, Simon J. Greenhill, Stephen C. Levinson, Russell D. Gray) deny that different features are universally connected, and say that the correlations are simply coincidences. The SVO/preposition relationship looks firm today, but over time we would see there was no necessary, universal relationship.
To make their case the authors used Joseph Greenberg's 1963 classic Universals of Grammar and drew up a list of proposed word order links. For example, if a the word-order of object and verb can be expected to correlate with the position of adpositions, [either a pre- or a post-position] the genitive-noun order (Joe's book v book Joe's or maybe even book's Joe), and relative clause-noun order (the book + which I bought v which I bought + the book).
The genitive-noun order catches my attention. At first, it sounds okay to me. French uses noun-adjective while English uses adjective-noun, but both use genitive-noun phrase and French, like English, puts the object after the verb. (My + blue book; mon + livre bleu). Thinking a little more, however, I recall that Swahili too uses verb-object, but puts the genitive last (kitabu kiblu + changu). So, the word-order rule cannot quite be universal, but then I know very few languages and there are about 7000 living ones plus many more extinct ones. In linguistics rare exceptions may challenge a rule without overthrowing it. Irregularities are part of the game.
The authors of the Nature letter sampled 625 languages from four language groups (Indo-European, Bantu, Austronesian, and Uto-Aztecan). They were not content, however, to simply to take a look at the languages and note the relationships. If that was all it took to settle the matter, it would have been settled long ago. Instead, they did an enormous analysis, pairing each language from a family, for example Albanian uses postpositions and object-verb order while Italian uses prepositions and verb-object order. Both Italian and Albanian are Indo-European languages, descended from the same common language, so at least one of the languages has changed these basic word-order rules. The authors then made two calculations. First, they figured the likelihood that the current word-order rules of each language evolved independently of one-another, and then the likelihood that they evolved together.
They had two hypotheses that should hold if the parameters for universals truly exist:
- Rules that are dependent on the same parameter should evolve together across the spectrum of language groups.
- No rules that depend on different parameters should evolve together.
The analyses found that neither hypothesis holds true.
The first fails spectacularly. There are no (0) relations that evolve together across all language groupings. Something as plausible sounding as a dependent relation between verb-object and noun-adposition order does not hold across the board.
The second fails as well. It turns out there are dependent relations in different languages (although the dependent relations are not the same in each family). It is not that there are no dependent relations between structural rules; it's just that the relations depend on the language family.
For this blog, the finding comes down sharply on the side of co-evolution and against the evolution of a linguistic module in the brain. It confirms very neatly the path that other research has highlighted.The story of speech origins is not a story of the origin of syntactic universals, or so says this latest work.
P.S. – So is this the final defeat of universal grammar? I'd like to say yes, but that is premature. This is the second major result reported in the past month dependent on mountains of computer-powered analyses using Bayesian statistical methods. It is plain that a great increase in statistical power and technique is underway and already yielding remarkable results in the study of language history. My temptation is to go with the results because the results support positions I already favored, but it looks as though I will have to knuckle down and learn a bit more about Bayesian statistics, not so much that I can do creative work in the field but at least so I can think a bit more critically about what these investigators are claiming. Anyway, it seems clear what statistical work Chomsky grumbled about in the panel mentioned in my last post. So far this work is not supporting his approach.