In his recent post on this blog, Simeon Floyd takes exception with my characterization of his 2011 Linguistic Typology paper as “asking a wrong question”. He emphasizes that the main point of his paper is to clarify what Quechua is like, not how to call the categories, and that in general it is not appropriate to accuse field workers (including field workers of the calibre of Marianne Mithun and Nicholas Evans) of universalism or even ethnocentrism, as I supposedly did in my 2012 paper on major word classes.
I’m glad that this kind of lively discussion is taking place on Diversity Linguistics Comment, as it confirms my hunch that there is a need for review pieces on articles, not just reviews of books. The blog medium seems quite appropriate for such informal discussion of more formal research papers (or overview papers, like my 2012 paper). I hope that more people will engage in similar discussions in the future.
I fully agree with Floyd that data accuracy is the foundation of everything we do, and there can be no serious theoretical and comparative linguistics without the kind of painstaking language-particular study that his 2011 paper on Quechua adjectives exemplifies well. I have great admiration for this kind of work, and I think it’s very important that descriptivists and comparativists learn from each other and respect each other’s approaches (and “if everyone does a bit of both”, as Floyd rightly says in a later comment). It seems to me that the growing interrelationship between the two groups of people has led to enormous progress in linguistics over the last few decades. The very term “diversity linguistics” is intended to capture the kind of symbiosis that we are increasingly seeing.
However, what exactly the relationship between fieldworkers and comparativists should be does not seem to be fully clear in the field. Many people, including fieldworkers, still seem to presuppose that there is an established set of categories with which language describers should work. This would lead to an asymmetric relationship between “theorists/comparativists” and describers: The latter have to work with the categories set up by the former. In contrast, I envisage a more equal relationship between comparativists and descriptivists: Each have their own set of concepts, research goals, and adequacy standards. Only comparativists are strictly dependent on descriptivists (because they rely on their data), while descriptivists can of course be inspired by the work of comparativists.
When I said that Floyd (2011) asked a “wrong question”, I did not mean the question whether adjectives are universal, but the question whether nouns and adjectives are distinct major classes in Quechua. As Floyd himself acknowledges in his comment: “you can call Quechua adjectives ‘a subclass of nouns’ if you must, or you can call them ‘class Z’ if you like, or you can call them ‘wugs’ “– how to call a language-particular category is not an important issue, and whether a category is a subcategory of some other category is not important either. What matters from the descriptive point of view is how the items behave, and what matters for the comparativist is whether they conform to a proposed generalization.
Floyd complains that typologists sometimes “squeeze a language into a pre-determined ‘type’ in support of some argument”, but I would contend that this is the bread and butter of typology. By their very nature, the typologists’ comparative concepts gloss over many subtle differences between languages – we simply cannot compare languages with respect to many different criteria simultaneously, and we have to limit ourselves to a few salient ones. It is undisputed that Quechua property-words behave like Quechua thing-words in that they occur in referential function without additional marking (e.g. hatun-ta rikaa [big-ACC see.1SG] ‘I see a big one’), and also in that they take a copula when occurring in predicative function. This is undisputably different from other languages such as Lezgian, which requires a special substantivizer on the noun (e.g. č’exi ‘big’, č’exi-di ‘big one’). Typologists who choose to focus on this difference between languages cannot be accused of misrepresentation – they just limit themselves to some key property. What Floyd does in his paper is add a lot of very interesting additional information, which typologists could eventually also take into account in some way. But these additional facts do not invalidate the original classification of Quechua.
Floyd notes that current typological work shows “how almost every point of comparison conflates cross-linguistic and language-specific morpho-syntactic criteria to some degree”, but he does not want to dismiss this work. By contrast, I tend to be very critical of typological work that conflates these kinds of criteria. For example, I have argued (Haspelmath 2011) that there is no good cross-linguistic definition of “word” and “affix”, and when people say that this doesn’t matter because there are language-particular criteria in each individual language, I am not satisfied. Different criteria in different languages single out different entities, and these cannot be used for typology.
Now I admit that it may seem like “a bit of a stretch to lump Floyd’s language-specific understanding of ‘adjective’ with Sandy Chung’s efforts to map the property of Chamorro word classes onto Baker’s universal word categories” (as Lev Michael notes in his comments on Floyd). But a generative universalist like Chung would be very happy with Floyd’s results about Quechua. Both Chung and Floyd use language-particular criteria (very different ones, as Chamorro is very different from Quechua) to argue for a general conclusion, namely the conclusion that lumpers like Schachter and Hengeveld (and Topping) are not right. My point is that both splitting and lumping is equally justified: A describer must split to be faithful to the language (as Floyd shows very well in his paper), but a comparativist like Schachter must lump. Now interestingly, the fieldworker Topping (who described Chamorro in 1973) was also a lumper. However, his point was not that there are no differences between adjectives and nouns (and intransitive verbs) in Chamorro. He just emphasized the similarities, because they seemed more salient to him, and he tried to highlight the differences between Chamorro and the familiar European languages. It seems to me that there has been a general shift of the mood among fieldworkers over the last decades: While previously they tended to emphasize the differences among languages, nowadays they tend to stress the similarities.
I hope that this clarifies the thrust of my 2012 paper a little bit. I do not think that the two positions are very far apart, but of course I understand that one reacts negatively if someone says that they “asked the wrong question”. What I wanted to say was that while work like Floyd’s is highly valuable and indeed foundational for typological work such as Schachter’s or Hengeveld’s, I do not see the two as contradictory. Maybe the original sin of typologists was to use the terms “noun” and “adjective” in a typological sense to begin with, instead of talking about property-words and thing-words. Using different terms in typology may be the best way to avoid misunderstandings.
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1). 31–80.
Haspelmath, Martin. 2012. How to compare major word-classes across the world’s languages. In: Graf, Thomas & Paperno, Denis & Szabolcsi, Anna & Tellings, Jos (eds.) Theories of everything: in honor of Edward Keenan, 109-130 (UCLA Working Papers in Linguistics, 17.) Los Angeles: UCLA.
Floyd, Simeon. 2011. Re-discovering the Quechua adjective. Linguistic Typology 15:25–63
Topping, Donald. 1973. Chamorro reference grammar. Honolulu: University Press of Hawai’i.
Hi Sergio, thanks for your comments. I’d just like to note that I did address frequency issues to some extent in my 2011 article (p50-58). In a medium-sized corpus I found about 4% of total adjectives used to head noun phrases, all cases with clear antecedents for ellipsis. When I manipulated the elicitation format with a task meant to provide clear antecedents from the context, this figure went up to 74%, highlighting how Quechua adjectives, unlike Quechua nouns, are highly context dependent in their possibilities for heading noun phrases. On English, Günther 2013 says quite a bit about the different construction types (with ‘one’ and without) and shows that full noun ellipsis is more common than our first impressions might suggest.
Oops, I meant to reply to Sergio’s comment below.
Yes, I noticed that.
I have thought about word classes (specificially in my favorite language family, Cariban) for some time now. Like everybody else, I am struck by the frequency with which claiming the existence or non-existence of a certain class boils down to prioritizing certain criteria over others (plus the eternal question, “but is it a major class or a subclass of X?”). Since this is quite analogous to the dialect vs. language discussion (cf. the parallel I made with dialect chains), my impression is that this question will never, can never be fully resolved. (Cysouw has proposed all those terms — doculects, linguoids, etc.; I suspect something analogous should happen in the study of word classes, i.e. new terminology that doesn’t really distinguish between the traditional “major classes” and “subclasses” but just names classes of whatever size. Perhaps one should also strive for a “categorial tree/network” showing the hierarchical relations between our various “classoids”, similar to networks used to represent dialectal chains, rather than the classical system of major classes and subclasses…)
“you can call Quechua adjectives ‘a subclass of nouns’ if you must, or you can call them ‘class Z’ if you like, or you can call them ‘wugs’”
What we call any morphosyntactic category in a language of course is unimportant, but whether a given category is a subclass of another category is of vital importance. Precisely the danger of ‘adjectives’ is that they efface this difference. We treat transitives and intransitives as types of verbs (and not as unrelated beasts), but many people treaty shizuka ‘peaceful’ and wakai ‘young’ in Japanese as two types of ‘adjectives’. In fact, the first is a type of noun and the second is a type of a verb. The claim A ⊂ B is an important and falsifiable claim for any two form classes in any one language, irrespective of how they are named. If things called ‘adjectives’ in Quecha are a subtype of things called ‘nouns’ in Quecha and Floyd did not say this, then this was his oversight or his mistake.
Unfortunately, the question of category membership is not as simple as your set-theoretical claim would have us believe. Criteria for defining the category to which a given word belongs can give contradictory results, or apply only marginally, or yield results that are a matter of degree. Such is the nature of the beast. Whether or not “adjectives” exist in Quechua, as an independent category or as a subcategory of nouns, is a debatable question (as is also the question of whether there are adjectives in Japanese, or nouns and verbs — a few google searches will yield articles defending opposing views on this question).
Yes, the problem is that cross-classification is rampant, so there is no simple hierarchy of categories. Unfortunately, we tend to think that categories have an essence that is independent of the properties, but they don’t. So we need to rank the properties, if we want to have classes with names that we can remember. Otherwise we’ll have thousands of categories which we can just number.
Nathan Hill: “If things called ‘adjectives’ in Quecha are a subtype of things called ‘nouns’ in Quecha and Floyd did not say this, then this was his oversight or his mistake.”
You might choose to analyze Quechua adjectives as a sub-class of nouns, but you would have to accept the consequences: Now you have added the adjectives to your noun class, and you suddenly have to explain why most of the nouns cannot, for example, be modified with degree words or reduplicate for intensification, or do any number of other adjective-like things. That is one reason I don’t opt for this account, but if you want to try to support it, you could attempt to analyze it that way. You will likely end up with a ‘nominal subclass’ and an ‘adjectival subclass’, at which point it makes little difference if you posit them as one class or two classes.
I don’t know if anyone is still reading these comments, but, for all it’s worth, here’s my USD 0.02.
I don’t think that the “real question” is so much whether Quechua has adjectives or not, whether they are an independent class by themselves (call them ‘adjectives’, ‘wugs’ aliaque curiosa) or “simply” a sublcass of nouns, etc., but really whether or not Quechua and English are meaningfully different.
Floyd shows that the difference between Quetchua and English is not that English cannot allow adjectives to head noun phrases — the “bigs” and “smalls” in his examples show that English does allow this to happen without the (usually called “necessary”) dummy element “one” occurring at the end of the phrase. So the difference between English and Quechua here is not “categorial” (can do vs. no can do) but rather of frequency of use (how often do “bigs” and “smalls” occur in English, and under what circumstances, when compared to Quechua? And what about “big ones” and “small ones”?).
There is something funny about the cases of “bigs” and “smalls” in English, something not quite all right — as if they were nouns rather than adjectives (one could perhaps talk about “bigs” and “smalls” as being specific items on a menu, as in a McDonald’s restaurant; in this way, the pluralized, nounless use of “big” is dependent on a very specific situation, in which it is being, as it were, nominalized to refer not to anything big, but to that specific menu item. The situation is reminiscent to me of cases such as the “nominalization” of expressions like “three o’clock” to mean the client that comes at that time (“is this your three o’clock?”, said one psychatrist to the other, pointing at the guy reading Time magazine in the waiting room) — not the same kind of regular process as when one nominalizes “read” into “read-er” via the suffix “-er”, but something else, more opportunistic or situation-dependent.
So English is different from Quechua; and it is more meaningful to compare Quechua adjectives without head nouns to English one-phrases (“big one”) rather than to “opportunistic nominalizations” (“bigs”). The opportunistic nominalizations are more of a fringe phenomenon, which could in turn develop and become someday the usual way of dealing with non-adnominal adjectives in English, but not really yet.
More generally, the whole discussion on classes (“but is it an adjective or a subtype of noun?”) hinges, I think, on a reification of these categories. It seems to me that the “real” discussion is what the properties of each (sub)group are, whether or not they are major, and the extent to which they coincide in defining the same terms. In other words: in a given language, are adjectives in a language better seen as a major category (because they must obey an important number of rules [= have a number of properties] that coincide on selecting the same items), or as a minor subcategory (because the number of properties is not so impressive, or they are not so major, or they do not coincide sufficiently on selecting the same items) — in which case adjectives, in that specific language, would be more like a semantically defined class, like, e.g., motion verbs, or kinship terms (classes which often do have specific morphosyntacitc properties setting them apart from other verbs or nouns).
Wouldn’t it be more interesting to consider properties, and clusters of properties and their corelations within a language and cross-linguistically? Isn’t it more interesting to consider whether adjectives can inflect like nouns (“bigs”), whether there is a dummy (“big one”) etc. cross-linguistically, rather than worrying about the typology of the adjectival class, and especially than worying about whether language X “has adjectives” or doesn’t? Wouldn’t a typology of properties rather than a typology of classes be a viable alternative? Isn’t it more interesting to ask whether the property of inflecting property-words in English (“bigs”) is different in form and/or function and/or scope from the property of inflecting proerty-words in Quechua (“hatun-ta”) rather than asking if “adjectives” in English are different from “adjectives” in Quechua?
(I’m reminded of dialectology, in which sometimes one has good dialect boundaries where the isoglosses coincide, thus delimiting a neat geographic area whose linguistic variety deserves the name “dialect X”, vs. dialect continua in which the isoglosses do not coincide and one has smooth transitions between near-identical varieties with extremes that are remarkably different, the so-called dialect chains. Dialectologists would say that naming a dialect as “dialect X” is useful in the former case — a number of different properties/isoglosses will be different in “dialect X”, so naming it facilitates the discussion — but not in the latter, where it is easier to talk about the isoglosses and their distributions and interactions without actually naming specific geographic areas between those isoglosses as “dialect X, Y or Z”. Wouldn’t it be wise in typology to also sometimes do the same, and not insist on labels like “adjectives” as being always more useful than the properties that define them? Could it be that it’s typologists who are sometimes asking the wrong question, or rather missing the more interesting comparanda?)
Just a few thoughts, quickly jotted down at 4 o’clock in the morning… just in case someone finds them interesting.
Yes, Sergio, these points are very well taken. A typology of individual properties will be more specific and more informative than a typology of the bigger classes. This is a point that quite a few typologists have made in recent years (most forcefully perhaps in Bathasar Bickel’s work, who also uses the term “multivariate typology”). But I think the bigger classes will not go away easily. These terms are simply too deeply entrenched, and too useful for our everyday work. So it’s worth trying to understand what they mean. And at least Bill Croft and Kees Hengeveld have made some very specific proposals about universals that can be stated in terms of the bigger categories, so I still see a role for them (even though their scope is much more limited than is sometimes thought).
Simeon Floyd may well be right that “the smaller-scale model of typology in which field linguists compare detailed and diverse primary data sets (often in topical workshops) leading to syntheses (often in edited volumes) tends to result in more meaningful generalizations than do large-scale approaches based on secondary data”. And this certainly is part of diversity linguistics, which includes studies of any degree of granularity, from language-specific study to abstract typology. But I still think that if we want to do coarse-grained typology and compare languages at the word-class level (using terms like “adjective” and “noun”), we need to confine ourselves to the basic features of attributive, predicative and referential function. If Quechua indeed is like English in these respects, i.e. if “big” in referential use is just as constrained as in English, then there is no typological contrast between the languages on this parameter. But as hinted by Floyd, such coarse-grained parameters may not lead to the most interesting results.
Distinguishing the rivulets from the larger stream is certainly a worthy goal, but it is fraught with danger, since one may lose track of why exactly this or that property (attributive, referential, predicative) is more important than the ones that show up in “fine-grained” comparisons of specific datasets. Which is why I keep wondering if it is not a better idea to compare the properties themselves (attributiveness, referentiality, predicativity) cross-linguistically, thus demoting the role of terms like “adjective” in the typology. In other words, isn’t it sometimes more about how comparable these properties/parameters are cross-linguistically than about whether or not the associated categories of lexemes actually “exist” (cf. the parallel with dialect chains)?
Good question – why is one property more important than another one? To some extent, this is a subjective question – we may simply find some properties more interesting than others, but other colleagues may have different preferences. Perhaps the only objective criterion is whether predictions can be made on the basis of a comparative concept. And I do think that some predictions can be made on the basis of broad classes of noun, verb and adjective (e.g. word order generalizations, but also others of the sort seen in Croft’s and Hengeveld’s (and van Lier’s) work.
I thank Haspelmath for acknowledging the importance of descriptive work and for advocating for a mutually-beneficial relationship with comparativists (although perhaps finger-pointing about “wrong questions” is not the best way to encourage it). In this spirit, I don’t think it is too much to ask that typologists look beyond practical uses of descriptive terms to determine to what extent they are being used similarly across descriptions of different languages. On the other hand, it is the responsibility of descriptivists to provide enough data to determine this (and it is true that they do not always achieve this). My observations on Quechua should be understood with this purpose in mind.
Linguists writing on word classes had been relying on limited Quechua data, so it seems that they should approve when richer information is provided (and to be fair, since publishing the article I have actually been thanked by several typologists for correcting their use of Quechua data). Haspelmath misses the point when he says that “these additional facts do not invalidate the original classification of Quechua”; what the additional facts do is invalidate spurious contrasts of Quechua with other languages like English. As he points out, language-internally one can optionally ‘lump’ or ‘split’, but in comparing two languages, one cannot arbitrarily lump one and split another.
For decades linguists repeated the line that “one type of language has nouns and adjectives, while another type has only nouns”, frequently using English as an example of the first type and Quechua languages as an example of the second. They justified these supposed ‘types’ with one main criterion: the ability of property words to (in certain contexts) head noun phrases. If, as Haspelmath says, “what matters for the comparativist is whether (items) conform to a proposed generalization”, this generalization simply does not hold up.
To illustrate this, we can consider the WALS question under which Quechua is classified as language in which ‘adjectives’ can occur without ‘nouns’, in contrast with its coding of English as allowing ‘adjectives’ only when ‘marked by following word’. This is clearly not accurate for many English constructions, both major and minor, frequent and rare (e.g. this sentence), so the generalization does not reflect the true similarities or differences of Quechua and English. In defense of this artificial contrast, some may appeal to ellipsis and lexical conventionalization to explain away the complications, but as I pointed out in my article, the same issues apply equally to Quechua. Quechua example sentences like those often cited as evidence in such arguments are actually very marked and cause native speakers problems of interpretation, and yet this problem is never addressed by those using Quechua data.
These basic issues of ecological validity tend to be ignored (after all, Quechua is that ‘type’ of language anyway, so there is nothing to explain), but these fundamental inconsistencies in the treatment of lesser-known unwritten languages with respect to standardized majority languages gloss over important issues of context and register. Comparing an idealized version of English with more conversational usages of Quechua is not helpful either for understanding how Quechua and English differ or how they resemble each other. I made similar observations already in the original article:
“Typologists should be skeptical of comparisons that ignore ellipsis, because if any elliptical constructions can be considered fully “grammatical” outside of the contexts that they depend on, then virtually nothing can be considered to be syntactically obligatory and much grammatical comparison becomes meaningless.”
“(Description should) be accountable to natural speech data, since grammaticality judgments in elicitation sentences cannot tell if a form is attested in normal speech, if it is relatively frequent or rare, or if it is restricted to specific contexts.”
I don’t think that a flippant attitude towards the complex facts is what Haspelmath is referring to when he says that applying pre-determined cross-linguistic ‘types’ on languages is the “bread and butter” of typology. Add the resulting spurious contrasts to the widespread basic inaccuracies in most databases (if the WALS coding for Imbabura Quechua, for example, is representative) and one might ask just how many of such truly “wrong questions” answered with “wrong data” can simply be written off as ‘noise’.
As an alternative, the smaller-scale model of typology in which field linguists compare detailed and diverse primary data sets (often in topical workshops) leading to syntheses (often in edited volumes) tends to result in more meaningful generalizations than do large-scale approaches based on secondary data; while their samples are much larger, their treatment of them is much more simplistic and sometimes inaccurate. I hope this more fine-grained approach is part of the “diversity linguistics” Haspelmath imagines, taking real natural speech data seriously, as I see it as one way that the barriers between the descriptivist and comparativist enterprises can be broken down.
I tend to agree with your perspective, and I am certainly a fan of the “edited-volume-from-workshop-on-richer-dataset” approach, which is my personal favorite way of doing typology. But I would say that the most important thing in this approach is the development of criteria that might be then more easily applied to other languages — so that, from the edited volume with 10-15 individual languages, a methodology arises that can be more massively applied to a larger number of languages, in the way that more traditional typologists have done. In other words, rather than a contrast between “megatypology” (lots of languages, not much data per language, cavalier attitude to data quality) and “microtypology” (few languages, lots of data per language, careful attitude to data quality), maybe a discussion of methods and how to discover them, so that they can be profitably applied to whatever range of languages/data the researcher happens to be interested in?
Consider, for example, the Quechua vs. English case. It strikes me that one criterion for investigating the differences between these languages, aside from discussing “ellipsis” or lack thereof, would be textual frequency. Yes, English can do things like “bigs” and “smalls” and questions big and small, but how frequent are such cases in a representative corpus of English? Similarly, how frequently do Quechua adjectives occur without a nominal head in a similarly representative corpus from one or more Quechuan languages? If this usage is, as the traditional typology suggests, more natural in Quechua than in English, then it should also occur more often.
I note, in passing, that there are other languages in which adjectives can occur unaccompanied by nominal heads: my own native language, Brazilian Portuguese, allows that to happen, and quite frequently (“o bonito”, “o grande”, “o pequeno”, etc.), without this having led to claims about the non-existence of adjectives. I suspect that a comparison of the frequency of such constructions in Brazilian Portuguese, Quechua, and English would also be an interesting exercise.