In a 2017 discussion note (in the Journal LT), Randy LaPolla confesses that he was “shocked” when he heard that some comparative linguists base their comparisons on the phenomena found in languages, not some underlying “causal” level of language (2017: 553). For example, in classifying Mandarin Chinese as an SVO (or A-V-P) language, they make use of the comparative concepts A and P, rather than the notion of Mandarin Topic, which is required to state the rules of Mandarin word order (and is in this sense a causal factor determining word order in the language).
This reminds me of a shock that I experienced 20 years ago, and that influenced my subsequent thinking: It was when Frederick Newmeyer published a paper with the title “The irrelevance of typology for grammatical theory” (1998). This old paper is not really news in 2019 (and Newmeyer expanded on his arguments in a subsequent book, Newmeyer 2005, which was read more widely), but it may still surprise some readers.
So could it be that comparative grammar and underlying structures or grammatical theories are not as intimately linked as is often thought? That comparative grammar is based on concrete “phenomenological” observations, while language-particular studies are interested in more abstract (or underlying) patterns, such as the mental grammars of individual speakers?
Newmeyer’s concern was primarily with “grammatical theory” in the sense of the innate prerequisites for language acquisition (“grammar blueprint”, or “UG”), and he has been arguing that Chomsky’s principles-and-parameters programme, which based its proposals about innate grammatical knowledge on typological evidence, was a failure (see also Newmeyer 2004); valid generalizations such as those discovered by Greenberg are instead due to functional-adaptive constraints of efficient coding (such as those proposed by Hawkins 2014). After my first shock about Newmeyer’s surprising claims, I found myself agreeing with this position: Innate grammatical knowledge, to the extent that it exists, does not seem to be responsible for the generalizations discovered by cross-linguistic research – very few if any of the proposals in this direction that were made in the 1980s and the 1990s seem to have stood the test of time (Haspelmath 2008).
But if the Greenbergian discoveries, and other universals found by subsequent research, are not attributable to the innate grammar blueprint, then there is actually no need to determine the mental grammars of speakers – it is sufficient to compare languages at the phenomenological level (e.g. using descriptions as we find them in pedagogical textbooks, which make no claim of mental realism). Functional constraints make speakers prefer certain patterns over others, and these need to be somehow mentally represented, but how exactly they are represented is immaterial. This is just as in other areas of culture: For example, certain settlement patterns (e.g. dense settlements with streets, central squares, and surrounding walls) are universally preferred by humans, but comparisons of towns across cultures and explanations of observed universals need not be based on studying people’s mental representations of town-building patterns. Just looking at towns around the world is sufficient.
Now LaPolla (2017) claims that such a phenomenological view is not sufficient: In addition to studying what speakers do, we need to ask what causes them to talk in a particular way, i.e. presumably their mental grammars. Or maybe their social grammars – i.e. the language-particular conventions that each speaker knows about (LaPolla does not specify which of the two he has in mind). For example, Mandarin speakers place the Topic early in a sentence, and they have no rule involving a “subject” notion (LaPolla 2009). I would not use the term “cause” here, but this is a terminological choice which is irrelevant here – what matters is that LaPolla is of course right that if we say that Mandarin is an A-V-P language (from a comparative perspective), then this is not based on the grammars or conventions of the speakers – it is based on a set of comparative concepts that were applied to Mandarin. Thus, there is a certain disconnect between language-particular grammars and comparative notions.
But why should this be a problem? It is widely recognized that token-based comparative concepts are often useful for comparison (i.e. questionnaire sentences, as pioneered by Dahl, or passages of parallel texts, as pioneered by Wälchli and Cysouw, or nonverbal stimuli, as pioneered by Stephen Levinson and his group), and these are clearly fairly distant from the mental or social rules (or “causes”) that the speakers are guided by. LaPolla does not really say why there should be a problem with phenomenological comparisons. He merely asserts that he thinks that
“the facts that explain the patterns found in the languages then should be operative in the comparisons, rather than the comparisons being done simply on a string of words with no reference to why the words are in the order they are.” (2017: 550)
The famous token-based comparisons do not fulfill this requirement, so why should comparisons based on grammatical descriptions fulfill it? Nobody would demand that comparisons of other cultural manifestations (e.g. town patterns) should be based on culture-specific rules (cf. the Japanese street address system, which is very different from Western systems and does not involve street names, even though Japanese towns have streets just as towns elsewhere). Comparisons in all areas of comparative culture research are based on special comparative concepts.
But more importantly, LaPolla does not say how one could compare languages in terms of “the (causal) patterns found in the languages”. Different languages have different conventions, and their speakers have different mental representations – what could comparison be based on?
In the traditional generative view, there is a set of universal features and categories that serve as a kind of “universal toolkit” from which grammars can choose, and which linguists are trying to discover. On this view, comparison should be done by means of these universal categories, which are also used for description. This is a coherent view (analogous to the chemical elements in the Periodic Table), and if we had some idea of what the universal categories are, we could proceed on this basis. But LaPolla does not share the generative view that description/analysis should be based on a hypothesis of what the universal categories are:
“we all agree that language description should be inductive and based on the facts of the language, that there are no crosslinguistic formal categories…, that language-specific descriptive categories used in describing a language are unique to that language” (2017: 548)
He says that “comparative categories are abstactions away from the actual details of the language-specific categories”, but this is often not sufficient: If one language bases its word order rules on a Topic notion, another one bases it on a notion of Nominative case, and a third one on a notion of Animacy, then the word order rules are thoroughly incommensurable.
Given this state of affairs, one strategy that comparative linguists have adopted is to free themselves from language-specific notions and look for other ways in which the languages could be usefully compared. One such approach appeals to the role-types A and P (agent and patient of physical-effect verbs and other arguments that are coded in the same way, Haspelmath 2011), and this has generally been seen as useful, especially since the order of the verb and the P-argument universally correlates with other word orders (Dryer 1992).
LaPolla may not be interested in this kind of comparison, but he has not identified any specific problem with it, so my preliminary conclusion is that the method (which goes back to Greenberg 1963) is sound, and I do not see any workable alternative to it (given that we are still very far from the principles-and-parameters hope that we will discover the universal categories of language). Thus, comparative linguists can work with special comparative concepts, and do not need innate universal categories (Newmeyer’s “grammatical theory”) or language-particular descriptive categories (LaPolla’s “causation”). This may be shocking news, but it is good news, because the other two kinds of concepts do not seem to provide a workable basis for grammatical comparison.