Hierarchical structures, and linguistic wars

Via Norbert Hornstein’s blog, I came across a recent paper in Royal Society Proceedings B (Frank, Bod, and Christiansen, 2012). In this post, I will present my own two cents on the new “language wars”, focussing on Frank et al.’s arguments against overestimating the roles played by hierarchical structure in language use, and on Hornstein’s treatment or lack thereof as exemplifying the defensive reactions generative linguists often display towards any and all challenges from outside. I will leave the debate on a certain Amazonian language to others.

The contested paper focuses on the psychological processes involved in language use and makes few definite statements about whether hierarchical structure is useful in describing the resultant structures of language, as when they state that “it is beyond dispute that hierarchical structure plays a key role in most descriptions of language. The question we pose here is: How relevant is hierarchy for the use of language?” (Frank et al., 2012, section 1; emphasis in original).

Their proposal points to valid questions: Whether or not resultant linguistic structures are best described in a hierarchical model, what is the empirical evidence that humans producing and interpreting everyday utterances are routinely making use of hierarchical operations? How use “cheats” to interpret utterances directly through the linear makeup of strings? Indeed, this type of questions – how simple can the underlying mechanisms be that we have to posit to explain the complexity of human languages? – are very much what a lot of linguists are on about these days. I remain very much unconvinced that linear sequence is the whole story. Here, Hornstein offers a few good counterarguments and difficult examples (in the few paragraphs where he refrains from pure mockery). What’s more, I suspect that the alternative model Frank et al. sketch, where potentially arbitrarily complex sentences are derived by jumping between parallel linear streams (as illustrated below), is anything than an opaque notation of hierarchical structure once you make it explicit enough to treat real data with it.

After all, what we minimally need to describe grammars of human languages with parallel streams, or parallel linear chunks, is some form of diacritic that tells the processer at which points it is legit to skip to the next stream, and which streams it can switch to. I strongly suspect that the parser doesn’t care whether you tell it “this is where you can switch to stream Y” or “this is where subtree Y is merged”, that these are possibly even mathematically equivalent descriptions of what’s underlyingly one and the same model. They are right to state that “any individual demonstration of hierarchical processing does not prove its primacy in language use” but they may have to be more careful not to fall for the trap of relabelling what is effectively a hierarchical structure under their model too.

A further problem I see is how they toss in recursion without prior discussion in their conclusions, when they state “even if some hierarchical grouping occurs in particular cases or circumstances, this does not imply that the operation can be applied recursively, yielding hierarchies of theoretically unbounded depth” (Conclusions). As argued in more detail by Martins (2012) and illustrated in his Venn diagram below, recursion, i.e. the possibility to embed an element within a larger element of the same type, does not per se imply unbounded iterations.

Examples of structures produced by iteration, hierarchical embedding and recursion, and by the combination of these processes. Constituents represented with the same letter are perceived as similar regarding a certain feature of relevance to the hierarchical structure. Brackets mean embedding. ‘ALPHA’ represents a general category that includes all letters in the alphabet. (Figure 1, Martins 2012)

My final contention is about how Frank et al. (2012) try to forge an argument for the primacy of sequential processes out of evolutionary considerations:

[L]anguage is shaped by constraints inherited from neural substrates predating the emergence of language, including constraints deriving from the nature of our thought processes, pragmatic factors relating to social interactions, restrictions on our sensorimotor apparatus and cognitive limitations on learning, memory and processing [21]. This perspective on language evolution suggests that our ability to process syntactic structure may largely rely on evolutionarily older, domain-general systems for accurately representing the sequential order of events and actions. Indeed, cross-species comparisons and genetic evidence indicate that humans have evolved sophisticated sequencing skills that were subsequently recruited for language [25]. If this evolutionary scenario is correct, then the mechanisms employed for language learning and use are likely to be fundamentally sequential in nature, rather than hierarchical. (emphasis added)

I agree with their stated premise, but not with their conclusion. Our ability to process syntactic structures may very well largely largely rely on evolutionarily older and domain-general systems without implying that what we do in language is “fundamentally sequential in nature”. The conclusion only follows if you add another assumption, namely that those older and domain-general mechanisms do themselves not show any hierarchical structure, which is itself an empirical claim they seem to take for granted without argument, but which may not hold up to closer scrutiny. As I sketched in an earlier post, there’s a growing line of research on possible hierarchical structure in the cognitive processing of complex action patterns such as are involved in tool use and processing, which may very well have been in place before language as we know it, and become re-employed, offering early language a ready-made foundation of mental machinery to deal with hierarchical structures.

Finally, as promised a word on the defensive reactions of generative linguists to outside challenges. Sure, it’s sometimes sad to watch how little impact actual linguistic research makes on work on human language by practitioners of other disciplines. The subject matter, though is objectively at the crossroads of many different fields of research, and telling others who want to make their expertise bear on the matter to just fuck off isn’t helpful. If others aren’t reading our work, it may be partly our fault for falling into esoteric jargon and failing to point out possible implications for broader questions. And who are we to tell psychologists not to make claims about language when we ourselves are – at least tacitly – making claims about (non-linguistic) psychology all the time by blaming certain aspects of what we find in language on general cognitive principles or “interface conditions”, often without making explicit enough how those cognitive principles are supposed to look like as to enable others to search for their signatures outside of language?

References and blogs quoted:

Frank, Stefan L., Rens Bod, and Morten H. Christiansen (2012). How hierarchical is language use? Proc. R. Soc. 279:1747, pp. 4522-4531; published online September 12, 2012, doi:10.1098/rspb.2012.1741

Hornstein, Norbert (2012). “Three psychologists walk into a bar…”, October 5, 2012, Blog post at “Faculty of language”.

Martins, Maurício Dias (2012). Distinctive signatures of recursion. Phil. Trans. R. Soc. B 367:1598, pp. 2055-2064 doi: 10.1098/rstb.2012.0097

Pereltsvaig, Asya (2012) “The Pirahã Controversy — Part 1“, blog post at “Languages of the World”


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s