Saturday, May 19, 2012

Back to Babel?

I've just finished reading David Bellos's recent book on translation, 'Is That a Fish in your Ear?' (Penguin, 2011). It verged between an entertaining read and a frustrating one. Perhaps unsurprisingly, I found the book to be at its least entertaining and convincing when it touched on subjects in my own area of expertise, linguistics. On the other hand, aspects of the book – such as the story of the dragomans of the Ottoman Empire (chapter 11), and the paradoxical language policies of the European Union (chapter 21) – were fascinating, and well narrated.

This isn't meant to be a full review, but since I feel that ITaFiyE misinterprets linguistics and/or sells it short at more than a few points throughout the book, I decided to set the record straight on the open web, especially with regard to three particular chapters.

Chapter 6: Native Command: Is Your Language Really Yours?

This chapter, on what it means to be a L1 or L2 speaker of a language, starts off promisingly enough. Bellos correctly observes that the traditional term 'mother tongue' is misleading, since we learn our L1 just as much from our peers as from our parents. He then goes on to claim that 'communicative competence' is acquired 'between the ages of one and three' – but that the language learned during this period is not always the one that adult speakers feel most comfortable using. The example he cites for this is Latin 700–1700, which uncontroversially had no native speakers during this period, but which was used as a vehicular language for various purposes. Then comes an astonishing leap (p59):
But if a clear distinction can be made between the language learned from your mother and the language in which you operate most effectively for high-born males in Western Europe between 700 and 1700 CE, the very concepts of 'mother tongue' and 'native speaker' need to be looked at again.
Um, really? The distinction seems pretty clear to me. The muddying of the waters in Bellos's book starts with use of the nebulous term 'communicative competence', which does not enter into mainstream definitions of native speaker status, for which grammatical competence is far more crucial. This is a minor quibble. But the claim that high-born males operated most effectively in Latin for a millennium is an incredible one. It may well be the case that for 'formal speech and writing', as well as for 'diplomacy, philosophy, mathematics, science and religion', Latin was the language of choice. However, these high-faluting academic purposes constitute a tiny minority of our total language use. Is the claim really that these people spoke (as adults) to their parents and peers in Latin in everyday situations? The suggestion that they 'thought' in Latin is even more absurd. There's a long literature on the 'language of thought' and how closely it approximates the languages we hear spoken, but the idea that a 15th-century Dutch nobleman, say, would wake up and think Sum esurientem ('I'm hungry') does not enter into it. The evidence from vernacular written traditions in Western Europe also speaks against this assertion. From the very beginning of the period 700–1700, writing – even for academic and ecclesiastical purposes – began to be carried out not in Latin but in the local languages of the area. Alfred's great program of translation into West Saxon English (not mentioned in this book), or the monastery translations of Boethius and other such texts in the Old High German-speaking area, are prime examples. These do not indicate that Latin was a language of thought, or even an effective operating language. Instead they indicate that the Latin of the period was a language on life-support, for which cribs had to be devised so that keen young men could get their heads round it. The worst part of this little paragraph is that even if it were true that a distinction could be drawn between 'the language learned from your mother' (read: first language) and 'the language in which you operate most effectively' in this instance, it wouldn't follow that this somehow invalidates the concept of a 'native speaker'.

This distinction continues to be made throughout the chapter, with the implication that languages learned during the early years of life are of little importance. There is, no doubt, a difference between 'first learned language', in Bellos's terms, and 'operative language'. Bellos then adduces two examples – his father, whose mother spoke to him in Yiddish but who learned English at school age, and his wife, who initially acquired Hungarian but who began to learn French at the age of five. The aim seems to be to deny the significance of the 'first learned language' or 'mother tongue'; and in these terms, it's a reasonable aim. But it misses the point that linguists and specialists in acquisition are trying to make when they talk about something called the 'critical period' or 'critical threshold', a term dating back to Lenneberg (1967). Very simply, in Trudgill's (2011) terms:
Lenneberg’s term refers to the well-known fact, obvious to anyone who has been alive and present in normal human societies for a couple of decades or so, that while small children learn languages perfectly, the vast majority of adults do not, especially in untutored situations.
To be sure, there is disagreement about what the relevant age is, or whether the term 'critical threshold' is really appropriate as opposed to a gradual tailing-off of language learning abilities. Meisel (2011) provides a recent summary. But what is uncontroversial is that adults do not learn languages as well as children. If, indeed, it is possible to isolate a specific age at which language learning ceases to be a cake-walk, that age is more like 7 (Meisel 2011: 134) rather than 3 as proposed by Bellos. Both of the examples that Bellos gives, then, may be evidence that 'first learned language' or 'mother tongue' is not what is important. But neither is problematic for the idea that languages learned during the critical period are learned better than languages learned after.

This misrepresentation colours the rest of the chapter, including Bellos's conclusion (pp65–66), to the effect that it is not important for translations to be into the translator's L1:
[I]t would be futile to insist that the iron rule of L1 translation be imposed on all intercultural relations in the world without also insisting on its inescapable corollary: that every educational system in the world's eighty vehicular languages devote significant resources to producing seventy-nine groups of competent L1 translators in each cohort of graduating students. The only alternative to that still utopian solution would be for speakers of the target languages to become more tolerant and more welcoming of the variants introduced into English, French, German and so forth by L2 translators working very hard indeed to make themselves understood.
There are a few things wrong with this. First, there's a straw man hiding amidst the prose. If we don't accept L1 translators, do we really have to devote huge amounts of money to training enormous numbers of language professionals? (Not that that would be a bad thing, in my opinion.) Not at all, because of something that Bellos himself mentions later in the book: translations can perfectly well be carried out via other languages. To find a translator from Welsh into Cantonese may be tricky, but when the translation passes via English it's a piece of cake. Of course there's a loss of fidelity involved in this two-step process – but one of the most convincing parts of Bellos's overall thesis (see chapter 10) is that the very notion of 'fidelity' is suspect when applied to translation anyhow.

More problematically, it seems no less 'utopian' to believe that L2 translators 'trying very hard' is the right solution. I'm not a translator, but I've worked in the industry, and my father's been in it for thirty years, so I feel I know enough to comment. 'Trying very hard' may be good enough when it comes to translation on a hypothetical academic-philosophical level, or literary translation (addressed in chapter 27 of Bellos's book but assumed tacitly throughout in the examples used, but in the real world of translation mistakes can be deadly. When I was working in Aachen, translating patient information leaflets from German into English, I was acutely aware that if I made a mistake people could be killed. With that in mind, it seems daft to insist that well-meaning L2 translators are as good as the real deal.

(There's also at least one factual error in this chapter. It is stated that 'all babies are languageless at the start of life'. That's not quite true: in fact, the process of language acquisition begins well before birth, as shown in experimental work by Kisilevsky et al. 2003 among others.)

Chapter 14: How Many Words Do We Have For Coffee?

Unlike chapter 6, this chapter sets out to argue for something reasonable. Its aim is to assess the evidence for linguistic relativity – the idea that language shapes thought – and its conclusion (in stark contrast to the gushing quasi-religious masturbatory rhetoric we so often see in the popular press surrounding the issue, for instance Boroditsky 2010) is sensible (p170).
If you go into a Starbuck's and ask for 'coffee' the barista most likely will give you a blank stare. To him the word means absolutely nothing. There are at least thirty-seven words for coffee in my local dialect of Coffeeshop Talk ... You should point this out next time anyone tells you that Eskimo has a hundred words for snow.
This is in general a strong and interesting chapter, even though the more recent work of 'neo-Whorfians' like Boroditsky and Levinson in the last decade is rather unaccountably left out of consideration.

My problem with it is only in how it begins: Bellos trots out a well-worn passage from Sir William Jones's 1786 Discourse, commenting (p161) that this 'is generally reckoned to be the starter's gun' in the development of comparative linguistics. The idea that Jones had this pivotal role is part of the origin mythology of historical linguistics, to be sure – but his significance has almost certainly been massively overestimated, as shown in detail by Campbell & Poser (2008: chapter 3). You can read their chapter if you want the real story, but the gist of it is this:

Firstly, Jones was not particularly original in his contributions. Commonalities between Indo-European languages had frequently been observed before his time, and even the relationship of Sanskrit to these languages was not a new idea.

Secondly, Jones made a lot of mistakes. He considered Peruvian, Chinese and Japanese languages to be part of the same family as the more familiar Indo-European languages, for instance, while leaving out others he should have included, such as Pahlavi, which he classed as Semitic (Campbell & Poser 2008: 37–38).

Thirdly, Jones was working within a biblical framework and viewed his own work as having 'confirmed the Mosaic accounts of the primitive world'; specifically, all the languages of the world could be traced back, according to him, to one of Noah's three sons Ham, Shem and Japhet (Campbell & Poser 2008: 40) – in stark contrast to the backbone of comparative linguistics of the day.

There's more to say, but the point should be clear enough. The idea that Jones was the founder of comparative linguistics is just as much of a myth as the idea that Eskimo has one hundred words for snow. The repetition of the myth is frustrating within the narrow confines of linguistics, and the situation can only get worse if books like this one, intended for a popular audience, perpetuate it further.

Afterbabble: In Lieu of an Epilogue

Epilogues are typically unambitious: summaries of the content and main argument of the book, perhaps, or suggestions for future research. Not so for the ITaFiyE epilogue, which tries – in 34 short pages – to solve the problem of language, the universe, and everything.

Well, perhaps that's overstating the case. But it does attempt to address the problem of the evolution of language, which is almost as thorny an issue. As modern theorists are fond of observing, in 1866 the Linguistic Society of Paris banned debates on the subject. Those same modern theorists often then argue that we have come far enough, nowadays, to lift the ban and talk about the origins of language with impunity. I disagree – though even among linguists I feel like I'm still in the minority here. We're barely any closer to understanding the genetic basis of our language capacity than we were a century ago, and there is still substantial debate as to what language even IS. The concrete proposals made by Noam Chomsky to that effect, as for example in Chomsky (1986), are very often rejected on the basis that they don't tally with our hazy pretheoretical intuitions about language – such as the idea that it is a social phenomenon, whatever that means; see e.g. Enfield (2010). We don't know nearly enough about human prehistory to say when language emerged, and that situation is unlikely to change. Most painfully, very often theorists still fail to distinguish between 'glossogeny', i.e. change in 'languages' (as we pretheoretically know them), and 'phylogeny', the emergence of the human biological capacity for language (whatever form that takes). (On this distinction, see Hurford 1990.)

In historical linguistics, meanwhile, it is quite normal to suggest that standard comparative methods typically can't take us more than 8–12,000 years into the past (Campbell & Poser 2008 have a discussion of this); this is not due to any flaw in the methods themselves, but rather to the build-up of confounding factors and the paucity of relevant data the further back one goes. Any claim about what was going on 40,000 years ago or more is likely to meet with extreme scepticism from any sensible historical linguist. Nevertheless, this is what specialists in the evolution of language get up to constantly. Perhaps not surprising that I can't shake the feeling that the whole field is a waste of time, then. Until someone has something more evidence-based to say, I'm inclined to take the simple route proposed by Berwick & Chomsky (2011): a tiny mutation emerged at some point, in one fell swoop, giving us the ability to put words together like we know we can; that mutation was (unsurprisingly) selectionally advantageous in the long run; and that's all there is to be said. (Curiously, critics of this 'saltationist' viewpoint are often the same people who rake Chomskyan linguistic theory over the coals for its apparent baroque complexity...)

But back to Bellos. He attacks the assumption that 'all languages are, at bottom, the same kind of thing, because, at the start, they were the same thing' (p341). Whether or not we believe the 'because' clause (and there's certainly no linguistic evidence that would lead us to; see again Campbell & Poser 2008), Bellos gives us no reason to doubt the underlying sameness of languages. The fact of linguistic diversity has very little bearing on this; the very fact that we have a concept of language at all, on the other hand, even a pretheoretical one, is evidence for sameness. At some level, we can judge whether something is linguistic or non-linguistic. That alone suggests unity in diversity.

Beyond that, though, there are linguistic arguments for sameness, many of which have been controversial over the years. Bellos groups these together as the argument that languages have 'a grammar', and disposes of it quickly and unconvincingly: rather than being an empirical matter, 'the "grammaticality hypothesis" is an axiom, a circular foundation stone' (p342). Why? Because...
Since traffic lights and the barking of dogs seem to have no discernible rules of combination or no ability to create new combinations, they have no grammar, and because all languages have a grammar in order to count as languages, dog barking and traffic lights are not languages. QED.
One might legitimately argue that this is hardly circular reasoning. Rather, we're attempting to understand the defining characteristics of language in terms of concrete properties, in order to establish what exactly it is that we're doing when we describe something as linguistic or non-linguistic. If these properties turn out to be a poor mapping to what our intuitive conception of language is, or not sharp enough to distinguish language from other things, we can reject them, refine them, retain them on the understanding that no better model has yet been proposed, OR decide that our concrete properties in fact tell us that our intuitive conception is wrong or not clear enough. This is what has happened over the last fifty years with Hockett's (1960) 'design features' of language (which seems to be what Bellos is bashing here, though he doesn't cite it), and with purported universals both in the Chomskyan tradition and the Greenbergian (e.g. Greenberg 1963). It seems to me to be normal scientific practice. For example, we now know that spiders are not insects, despite all our intuitions telling us otherwise, and this is a natural consequence of adopting a particular model of taxonomic classification that is superior to our vague intuitions. We now know that whiteness isn't necessarily a property of swans. Likewise, by the botanical definition of tomatoes that we adopt, they are classed as fruit rather than vegetables. This isn't just axiomatics: we're learning something that we didn't already know. This is scientific progress.

But Bellos, like many authors in and around linguistics, refuses to give up on the intuitive definition. He proceeds as follows (p342):
In a similarly circular way, the axiom of grammaticality pushes to the edge of language study all those uses of human vocal noises – ums, hums, screams, giggles ... and so forth – that don't decompose neatly into nouns, verbs and full stops.
Quite apart from the disingenuousness of this comment (no one ever seriously proposed that full stops were a property of human language, axiomatically or otherwise, as Bellos must well know), I fail to see the problem. Describing these types of noise as 'non-linguistic' seems to me to be entirely fair and reasonable. (Note that this is very different from saying that they don't constitute a worthy object of study in their own right, a claim that no linguist I know would want to be associated with.) What we discover by doing this kind of scientific work is that our intuitive conception of language is so fuzzy and all-encompassing as to be effectively unusable, a point that Chomsky has been making for years (see again Chomsky 1986). If, like Bellos, you find definitions in the Chomskyan mould unpalatable, then the onus is on you to come up with a better operational definition if you want to be thought of as doing serious work on language.

After listing various ways in which languages can be odd (evidentials always seem to come up whenever anyone puts together a list like this!), Bellos somewhat uncharitably (but perhaps not unfairly) states that the attempt to discern what all grammars share 'has got about as far as the search for the Holy Grail'. He then builds another straw man which he proceeds to rip apart: 'all grammars regulate the ways in which free items may be combined to make an acceptable sentence' (p344). The obvious problem is the word 'sentence' here – what does it mean? Nothing (either inside or outside a theory of grammar, as far as I'm aware), and of course we don't speak in sentences, as Bellos points out.

Nor is it really a problem that 'no living language has yet been given a grammar that accounts for absolutely all of the expressions' (p344). Even if this goal were a reasonable one for linguistic theory (Chomsky 1986 argues that it isn't), and even if a living language like 'English' were a coherent object of study (see virtually any work by Chomsky for a brief but irrefutable demonstration that it isn't), does this stymie any attempt? In physics, our best theories of reality can't account for phenomena like dark matter; all this shows is that science (any science) is a work in progress. So it's an absolute nonsense to claim, as Bellos does (p344), that:
Flaws of this magnitude in aerodynamics or the theory of probability would not have allowed the Wright Brothers to get off the ground or the National Lottery to finance the arts.
First off, there's no reason that our scientific theory has to be practically applicable in order to be worth something (look at string theory, for instance). But, in any case, Bellos should look at state-of-the-art work in computational linguistics, where parsers based on handwritten grammars in combination with a simple statistical learning algorithm can robustly parse up to 92.4% of an average corpus of English (see Fossum & Knight 2009). That doesn't seem like crash-and-burn to me.

The afterbabble goes on to compare dialectal variation to primate grooming, and to propose this as a potential evolutionary origin for language, following work by Robin Dunbar. I won't discuss this in any detail, but suffice it to say that the conclusion – that 'The most likely original use of human speech was to be different, not the same' (p351) – presupposes precisely what has been argued so vigorously against earlier in the same chapter, namely a definable object that is 'human speech' (which, since animals fairly intuitively don't have it, must have evolved somehow).

In short, this chapter (and the book as a whole) overreaches itself. Though issues of translation are inevitably bound up with deep questions about the nature of language, ITaFiyE would have been a better book if it had chosen to stick closely to the former and leave the latter to specialists.

References

Berwick, Robert C., & Noam Chomsky. 2011. The biolinguistic program: the current state of its evolution and development. In Ana Maria di Sciullo & Cedric Boeckx (eds.), The biolinguistic enterprise: new perspectives on the evolution and nature of the human language faculty, 19–41. Oxford: Oxford University Press.
Boroditsky, Lera. 2010. Lost in translation. Wall Street Journal, 23 July.
Campbell, Lyle, & William J. Poser. 2008. Language classification: history and method. Cambridge: Cambridge University Press. 
Chomsky, Noam. 1986. Knowledge of language. New York: Praeger.
Enfield, Nicholas J. 2010. Without social context? Science 329, 1600–1601.
Fossum, Victoria, & Kevin Knight. 2009. Combining constituent parsers. Proceedings of NAACL HLT 2009: Short Papers, 253–256.
Greenberg, Joseph H. 1963. Some universals of grammar with particular reference to the order of meaningful elements. In Joseph H. Greenberg (ed.), Universals of grammar, 73–113. Cambridge, MA: MIT Press.
Hockett, Charles F. 1960. The origin of speech. Scientific American 203, 89–97.
Hurford, James R. 1990. Nativist and functional explanations in language acquisition. In I. M. Roca (ed.), Logical issues in language acquisition, 85–136.
Kisilevsky, Barbara, Sylvia Hains, Kang Lee, Xing Xie, Hefeng Huang, Hai Hui Ye, Ke Zhang, & Zengping Wang. 2003. Effects of experience on fetal voice recognition. Psychological Science 14, 220–224. Lenneberg, Eric. 1967. Biological foundations of language. New York: John Wiley & Sons.
Meisel, J├╝rgen M. 2011. Bilingual acquisition and theories of diachronic change: bilingualism as cause and effect of grammatical change. Bilingualism: Language and Cognition 14, 121–145. Trudgill, Peter. 2011. Sociolinguistic typology: social determinants of linguistic complexity. Oxford: Oxford University Press.