The Piraha discussion continues

Via Language Log/John Cowan: Dan Everett's finally gotten around to publishing a few more examples of his claims about Piraha - notably, that they have no recursion, and in particular no subordinate clauses Even quoted speech and conditionals, he claims, are not embedded. Here it is: Pirahã culture and grammar: A response to some criticisms.

Now, recursion means being able to embed a given kind of phrase within another example of the same kind of phrase, as many times as you want. In "the door of the house", one noun phrase ("the door") is embedded within another one ("the door of the house"); in "I will visit you when it stops raining", a clause "it stops raining" is embedded within a larger one ("I will visit you when it stops raining"). You can also keep doing this ("the edge of the handle of the door of the house", "I will visit you when I know whether Khaled said that James is right about the forecast that it will rain tomorrow.") In Piraha, Everett reports that for noun phrases you can only do this once (no more than one possessor), and for clauses that you can't do it at all (he insists that all the examples that look like subordinate or adverbial clauses are actually separate sentences whose linkage is left for the listener to interpret, and in this paper presents some arguments for this.)

The thing is, a language with such properties has obvious potential to be expanded into a language like English or Arabic. For possessors, all it would take is a little analogical expansion - that's what allows us to interpret a phrase like "my brother's wife's cousin's friend's cat's teeth" as grammatical, even though you may well never have heard a noun phrase with six possessors before. For subordinate clauses, all it would take is grammaticalising some kind of erstwhile adverb or intonation pattern or quotative marker into a signal that these two clauses are more closely bound than others; such changes occur all the time in languages that already have subordinate clauses (eg "with what" > "in order to" in Algerian Arabic.) If the Piraha haven't done this, then why not? If they used to speak a language with multiple possessors and subordinate clauses in the past, why and how did they abandon these features - and if they never have, then why have most languages gained these features? In short, what motivates the expansion of grammar, and how does it happen?

One place (doubtless not the only one) where I think you can see expansion of grammar in action is technical terminology; consider mathematics. "The set of all p/q such that q!=0 and p, q are integers" is perfectly clear mathematical English, but is rather unlikely to be heard in everyday English (? "the set of all couples such that the husband is not an accountant and both the husband and wife are from Belgium"). The needs of mathematical communication have motivated the use of a kind of relative clause, with a complementiser and neither a gap nor a resumptive pronoun nor a relative pronoun, which is at best marginal in normal English; if enough people were trained as mathematicians, it might get used more widely. Maybe multiple possessors and subordinate clauses are technical features to cope with the demands of socialising with large numbers of people. Or maybe Piraha has a little more embedding than Everett reports. Speculation is fun, but a nice big, searchable, publicly available corpus would be a lot more convincing.

Open to interpretation

Songhay's lexical economy - the way it keeps its lexicon rather smaller than its neighbours' by using a single word to fulfill the functions of what in most languages would be several different words - has attracted the attention of several of those who have written about the language from the 1850s onwards. While Kwarandzyey (Korandje) is so full of Berber and Arabic loanwords that the size issue probably no longer applies, it still has many striking examples of polysemy. Take "open", for example.

fya (from Songhay *feeri) is best translated as "open" (its commonest sense). Of course, to open one's mouth can be to start eating - hence the frozen compound fya-mmi "open-mouth" means "breakfast". But opening is also what you do to release something from an enclosed space; hence to "open water (for something)" (fya iri), or just "open", is to irrigate, and to "open for an animal or person" is to release them. Likewise, to "open a rope (for something)" is to untie it. To release something from your grasp is to let it fall - hence to "open for something" is also to drop it. And for a man to release his wife from her obligations towards him is to end the marriage - hence to "open for a woman" is to divorce her.

We can map the connections between these easily enough, making it clear that they form a coherent network of meaning:

breakfast untie
    \    /    \
     open - release
       \      / \
       irrigate divorce

But not only will any single English translation applied literally and consistently yield ludicrous results for at least some of these cases - translating it differently in different circumstances will force you to choose a single meaning in cases where the text is ambiguous. "He opened for the woman" probably means he divorced her, but in principle it could mean he released her (eg from prison), or untied her, or (literally) dropped her; in fact, since Songhay has no gender distinctions in pronouns, it should even be able to mean "It (eg an automatic door) opened for her". And of course, this kind of ambiguity can be deliberately exploited for effect, as in puns.

In Kwarandzyey, this is never likely to cause serious ambiguity - the language is almost never written down, and it's a small enough community that the context is usually known to everyone anyway. But imagine worrying about this kind of thing in a millennia-old text in a language that no one today speaks natively, and you can really see why even the most literal translation of such a text is unavoidably an act of interpretation.

Why dead snakes are like clothes

What would you say if, in some science-fiction novel, you read of a language where the situations that in English would be described as "The clothes blew down from the clothesline", "Push that dead snake away with a stick", and "I see where he's carrying the rabbits he killed hung from his belt" were all naturally expressed with the same root, plus nothing more than different affixes? What about "I slammed together the hunks of clay I held in either hand", "I slung away the rotten tomatoes, sluicing them off the pan they were in", and "I picked up in my mouth the already chewed gum from where it was stuck on the table"? My inclination would have been to dismiss it as a neat but implausible idea, placing some strain on the reader's suspension of disbelief. But - until no more than thirty years ago - such a language existed right in California. Go to Part III of Leonard Talmy's dissertation Semantic Structures in English and Atsugewi to get the data; here's a slightly less surprising example as a taster:

s-'-w-	cu-	lup-	hiy-ik:-	a
Subject=I, Object=3rd person	from a linear object moving axially [with one end] non-obliquely against the FIGURE	for a small shiny spherical object to move	out of a snug enclosure/a socket	factual
I poked his eye out (with a stick.)

s-'-w-	pri-	lup-	nik-iy-	a
Subject=I, Object=3rd person	from the mouth/interior of a person, working ingressively, acting on the FIGURE	for a small shiny spherical object to move	all about, here and there, back and forth	factual
I rolled the round candy around in my mouth.

Of course, people are people; after explanation, the similarities are easy enough to make out, and presumably given enough time anyone can learn to look at a situation and decompose it into elements like these, rather than the elements that "leap out" at an English speaker. In fact, I suspect that having to learn to see things the way the people you talk to do is one of the subtler drivers behind contact-induced language change. But cases like this provoke thought: just how much can the attributes of a situation most relevant to formulating a sentence vary from language to language?

Flora of the Central Sahara and elsewhere

Ever found yourself trying to sort out a plant name you've elicited, not knowing any botany worth mentioning? Well, it turns out the botanists are a step ahead of the linguists on the digital libraries game, at least in Spain: the Digital Library del Real Jardín Botánico CSIC has a pretty remarkable array of books to browse online. The one that just saved my etymology of the Kwarandzyey plant name tsifəṛfəẓ is Etudes sur la flore et la végétation de la Sahara centrale. Vol. III: Hoggar, which gives both Tamasheq and binomial names for each plant mentioned. Unfortunately it's clear that not all the works give translations of the names, but it's still worth a look.

On a similar note, I've found Sahara-Nature handy sometimes.

Colour vision and language shift

In a brief Edge article (see LH), Lera Boroditsky makes the thought-provoking remark - regarding perception of colours - that “It turns out that languages meddle in very low-level aspects of perception, and without our knowledge or consent shape the very nuts and bolts of how we see the world.” If this is so, what happens when pretty much every speaker of a given language is also fluently bilingual in another one which divides up the spectrum (or indeed the world) differently - as has been the case here in Tabelbala for at least two generations? As it happens, some of my recent work here points to an answer.

I've recently been examining the colour system of Kwarandjie, trying out the second half of the Berlin and Kay tests (focus identification) with a number of speakers (well, 13 so far.) Of course, like all speakers of Kwarandjie, they are bilingual in Algerian Arabic; in fact, many of the speakers tested speak Arabic better than Kwarandjie. The colours they see turn out to be remarkably consistent, with more or less the same foci from speaker to speaker: black, white, red, yellow, green, and blue (as well as some secondary colours, most commonly pink (Arabic wəṛdi or, in reference to a darker shade, ħənnawi), that are less widely agreed on.) However, the words used to refer to “green” and “blue” show significant variation. For some speakers, zəgzəg means “blue” and “green” is (Arabic) xḍəṛ; for others, zəgzəg means “green”, and “blue” is (Arabic) ẓərrig!

It doesn't require too much speculation to think up a scenario to explain this. A few generations back, Kwarandjie must have had a five-colour system, featuring (like Japanese aoi, for example) a colour zəgzəg which covered both green and blue, whose focus was somewhere between the two. As speakers grew more fluent in Arabic, this focus split; they came to see both green and blue. Depending on whether they more frequently heard older speakers refer to, for example, plants or the sky as zəgzəg, they decided it meant one colour or the other, and gave the other colour an Arabic name; but different choices were made in different families. In the coming weeks I hope to gather more evidence on the issue - in particular, to learn whether even older speakers than those examined see a single colour grue or not.

(Non-)universal quantifiers

Many readers will recall Everett's argument that Pirahã had no universal quantifier because statements featuring what he had originally translated as "all" would generally be considered true even if a small part of the original had to be excepted. I'm not sure the conclusion follows (universal quantification could still be its prototypical meaning, for example), but if it does, then it could equally well be argued to be true of Darja; a lot of statements about "all" that I hear made here are ones which the speaker is perfectly aware (and accepting) of the existence of exceptions to, and it took me a while to go against my mathematical training and realise that when they said "all", they didn't mean it in the logicians' sense. Actually, I suspect the same is true of many idiolects of English. This was brought to mind by a little example I heard yesterday: hađa ybi` kŭll ħaja, bəṣṣəħħ əlfṭayəs ma ybi`š هذا يبيع كُلّ حاجة، بصّح الفطايس ما يبيعش "This guy sells everything, but hammers he doesn't sell."

Impersonal vs. personal "you"

In English, "you" is equally used in a literal sense (referring to the addressee) or in an impersonal sense (referring to an arbitrary imagined experiencer.) In Darja, at first sight, it looks the same way - and for speakers of any one gender, this is true. However, looking at speakers of both genders allows you to realise that the distinction is grammaticalised. Addressee "you" agrees in gender with the addressee; impersonal "you" does not agree in gender with the addressee, but with the speaker. Thus a woman speaking to a man will say tṛuħ "you go" when "you" refers to the man addressed, but tṛuħi when it refers to an arbitrary person, like "When you go by bus, it takes a while."

The Piraha debate heats up

A recent Language Log post alluded below the fold to two very interesting papers continuing the Piraha debate. Piraha Exceptionality: a Reassessment (Nevins, Pesetsky, and Rodrigues) reexamines Everett 2005's claims in light of Everett 1986, pointing out substantial and inadequately explained discrepancies between the two, and concluding with the rather hard-hitting statement that:

CA asserts, for example, that the embedded clauses amply documented and described in the earlier work are not actually embedded clauses, but offers no account or even acknowledgment of the numerous facts that argue in favor of the old view over the new. Similarly, CA offers as an argument for the new view the absence of long-distance wh-movement, but offers no new account of the data that in earlier work motivated the claim that Pirahã has no overt wh-movement of any kind. Likewise, as we have seen, CA asserts that Pirahã lacks quantifiers, but offers no coherent evidence against the proposal that the words described as quantifiers in the earlier work were described wrongly. In section 5, we have suggested that the situation is little better with respect to CA's discussion of Pirahã culture. CA simply asserts that Pirahã grammar has properties that, if true, would place it outside the pale of grammar and culture as we know it and would demand a special explanation for Pirahã's seeming uniqueness.

Everett replies in "Cultural Constraints on Grammar in PIRAHÃ: A Reply to Nevins, Pesetsky, and Rodrigues (2007). He protests their efforts to provide comprehensible glosses for his 2005 sentences, objecting that considerations like what "the best free translation, the least exotic translation" is are irrelevant to the final analysis, which should rely solely on the truth conditions for the word's use, and that in any event "armchair linguists who wouldn't be able to pronounce a single Pirahã word" are in no position to give such glosses. Glosses like "cloth arm" are superior to glosses like "hammock" (which is what the compound in question means), because they help inform the reader about the complexity of Pirahã morphology. He also offers some interesting evidence on why he now analyses what he had previously termed a "nominaliser" (and had glossed as such in 2005) as a marker of old information. His core objection seems to be that such efforts as Nevins et al's are bound to fail because not all languages "translate fairly well into one another", and in particular, Piraha cannot be translated well into English; the comprehensible translations they propose don't have the same truth conditions, and the "literal" "translations" (yes, I think that was worth two pairs of scare quotes) that he sometimes gives (eg Everett 2005:624: “Smallness of cans remaining associated was in the gut of the canoe”; what would the truth conditions for something being "in the gut of the canoe" be, I wonder?) don't exoticise the language so much as attempt to render its genuine exoticism into English.

The debate looks like an argument about where the burden of proof lies: for example, does Everett need to provide more than two examples of how the truth conditions of Piraha "ba´aiso" differ from those of English "whole" (supposedly; the anaconda skin example works just fine for me in English, presumably implying that my word "whole" does not in fact mean the same as Everett's word "whole"), or do his critics need to go learn Piraha before they can question his claims about the meaning of "ba´aiso"? Are his critics justified in assuming that, in the absence of contrary published evidence, a given Piraha structure will have a familiar counterpart? Extraordinary claims require extraordinary evidence.

Incidentally, Everett's response provides another interesting example of differing truth conditions for a sentence in English. In his idiolect, apparently, the fact that, when outsiders come,

"They say hello and the Pirahãs say hello back. They ask if there are any fish and the Pirahãs say that there are fish or are not fish. Many Pirahãs can communicate at a rudimentary level in Portuguese. But they lose the gist of conversations very easily and often after someone has left they ask me to interpret... The best speakers of Portuguese among the Pirahãs speak it about as well as I do French. I can say a few things and find a bathroom, but I am not ready for any conversation of any depth at all."

is so perfectly compatible with all Piraha being "monolingual" that he can actually offer it as evidence for the claim.

Of truth and Scotsmen

A few commentators on my previous post invoked the "No true Scotsman" fallacy. This term refers to a rhetorical move whereby one makes a sweeping generalisation about some class (say, Scotsmen), and, when presented with a counterexample to the generalisation, retorts that this counterexample is not a true example of the class ("no true Scotsman"). While not strictly speaking fallacious, this move is of course misleading, in that it suggests that the empirically testable generalisation first proposed still stands, when actually it has been replaced with an unfalsifiable tautology applying to a much narrower class. I didn't in fact use any such argument in my last post, but I smelled something fishy about this "fallacy", and decided to examine it further.

The "true X" construction is semantically productive in a wide variety of contexts (a true friend, a true patriot, a true gentleman, a true villain, a true genius, a true cat...) In each case, it presupposes that some things that would normally be termed members of the class are not true ones (that not all of those who are normally labelled friends are true friends, for example). At least for animate nouns, it seems to distinguish between a broader use of a term based on appearance or convention, and a less inclusive one referring to a sort of ideal that members of the broader class may or may not live up to. Thus a true cat is one that exhibits, in exaggerated form, all the characteristics that we associate (accurately or not) with stereotypical cats; a true friend is one who exhibits the characteristics of friendship in circumstances where others would fail to exhibit them. Anyone can be a Scotsman merely by being born in Scotland, but to be a true Scotsman, you probably have to wear a kilt all the time, love haggis, roll your r's, etc.

Obviously, this feature of English is not inherently fallacious or misleading; using the phrase "true X" simply saves us the trouble of saying "an X that more than satisfies our expectations of Xs". Now that more people actually know foreigners personally rather than by report, it's become painfully obvious how far from reality most national stereotypes are, so this construction sounds rather silly when applied to Scotsmen; but even there it nonetheless conveys a fairly clear meaning to anyone familiar with English culture.

Interestingly, though, the rhetorical move originally criticised is not always misleading either. You can undeniably be a Scotsman (or a cat) without being a true Scotsman (or a true cat), so the suggestion that calling someone not a true Scotsman saves the original sweeping generalisation about Scotsmen is false. But if you're not a true friend (or a true gentleman), you're not really a friend (or a gentleman) at all - just as fool's gold isn't really gold - so calling someone not a true friend, if justifiable, does save your sweeping generalisation about friends, or, more accurately, your statement defining the characteristics of a friend. A poem I studied way back in middle school provides an entertaining example of the non-fallacious use of an (implicit) "No true Scotsman"-type argument:

Ramped and roared the lions, with horrid laughing jaws;
They bit, they glared, gave blows like beams, a wind went with their paws;
With wallowing might and stifled roar they rolled on one another,
Till all the pit with sand and mane was in a thunderous smother;
The bloody foam above the bars came whisking through the air;
Said Francis then, “Faith, gentlemen, we’re better here than there.”

De Lorge’s love o’er heard the King, a beauteous lively dame,
With smiling lips and sharp bright eyes, which always seemed the same;
She thought, The Count my lover is brave as brave can be;
He surely would do wondrous things to show his love of me;
King, ladies, lovers, all look on; the occasion is divine;
I’ll drop my glove, to prove his love; great glory will be mine.

She dropped her glove, to prove his love, then looked at him and smiled;
He bowed, and in a moment leaped among the lions wild:
The leap was quick, return was quick, he has regained his place,
Then threw the glove, but not with love, right in the lady’s face.
“By Heaven,” said Francis, “rightly done!” and he rose from where he sat;
“No love,” quoth he, “but vanity, sets love a task like that.”