2. On the Limits of the Comparative Method.pdf

2. On the Limits of the Comparative Method : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:28 PM

2. On the Limits of the Comparative Method

S. P. HARRISON

Subject

Linguistics » Historical Linguistics

Key-Topics

comparative , methods

DOI:

10.1111/b.9781405127479.2004.00004.x

In this chapter, I explore the limits of the comparative method as a tool in comparative historical linguistics. 1

Let me be quite clear about one thing from the outset: for me, the comparative method is the sine qua non

of linguistic prehistory. I believe that the comparative method is the only tool available to us for determining

genetic relatedness amongst languages, in the absence of written records. I believe that prior “successful”

application of the comparative method is a prerequisite to any attempt at grammatical comparison and

reconstruction. But the comparative method has limitations, determined by the very properties of the method

that make it work:

i It has relative temporal limitations . The more changes related languages have undergone (in

general, a function of time), the less likely the method is to be able to determine relatedness.

ii It has sociohistorical limitations . Certain historical situations can have linguistic consequences that

vitiate the comparative method.

iii It has linguistic domain limitations . Only certain sorts of linguistic objects can be usefully

compared and reconstructed using the method.

iv It has limitations of “delicacy.” Only genetic relationships up to a certain degree of precision or

delicacy can be reliably determined using the method.

I discuss each of these types of limitation in turn below.

Disagreements and misunderstandings regarding what the comparative method can and cannot do are a

continuing (and, some might say, distracting) leitmotif in comparative historical linguistics. The level of

disagreement has often surprised me, and must be attributed to some level of disagreement regarding what

the comparative method in historical linguistics actually involves, what its premises are, and what its

recognized argument forms are. My first task, then, must be to outline what I think the method is.

In section 1, I outline what I see as the goals of comparative historical linguistics. In section 2, I describe

how the comparative method serves to realize those goals. The limits and limitations of the comparative

method are treated in section 3. Sections 3.1 and 3.2 discuss the possibility of comparing and reconstructing

grammar, both with and without the comparative method. Section 3.3 discusses two circumstances in which

the comparative method may fail to recognize genetic relatedness. Section 3.4 is devoted to the unique

problems posed by subgrouping. Section 4 considers briefly how the comparative historical linguist can

survive the limitations on the comparative method.

1 The Goals of Comparative Historical Linguistics

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274794

Page 1 of 23

2. On the Limits of the Comparative Method : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:28 PM

Identifying the goals of comparative historical linguistics is not a particularly problematic exercise. They are

essentially three in number:

i to identify instances of genetic relatedness amongst languages;

ii to explore the history of individual languages;

iii to develop a theory of linguistic change.

Nor, of course, are these goals in practice independent. The identification of instances of genetic relatedness

is likely to be a concomitant of the investigation of the histories of one or more related languages. The

development of a theory of linguistic change is informed, one trusts, by investigation of the histories of

individual languages and language families.

Prehistorians might be satisfied with (or, at least, most immediately interested in) results stemming from the

first of these goals, and cultural historians with the second. “True” historical linguists view the third goal as

the real prize, the ultimate aim of the exercise. That is certainly how I rank the goals. I want to know

whether one can distinguish possible from impossible changes, or, at the very least, probable from

improbable. I want to know whether or not there are any constraints on borrowing. I want to understand the

engine of language change – how changes begin, and how they move through languages and linguistic

communities.

The desiderata of such a theory of language change were set out quite clearly over a quarter century ago in

Weinreich et al. (1968). Some aspects of the research program they outlined have been elaborated in

subsequent work. Labov and others have studied cases of language change in progress (cf., e.g., especially,

Labov 1994 for discussion and extensive references). The regularity assumption (see below) has been put

under scrutiny in their work, and in the work begun by Wang (1969; cf. also Wang 1977) on the so-called

“lexical diffusion” of sound change. The notions “natural linguistic process” and “natural linguistic system”

(and, derivatively, “natural linguistic change”) have been the focus of linguistic theory from the time

Weinreich et al. (1968) appeared. More recently, scholars like Sarah Thomason have given detailed

consideration to the limits of borrowing and diffusion. 2 But, we are still some distance away from a theory of

language change.

2 The Place of the Comparative Method

A theory of the sort envisaged in the preceding section is one that, given some synchronic language state S,

would tell us what immediate antecedent state(s) P S + could/must have given rise to S. Such antecedent state

sets for different languages could then be compared for “best fit,” in order to select amongst potential

antecedent state candidates (if the theory supplies more than one) and to determine genetic relatedness. In

the absence of such a theory, 3 however, the comparative method has served the historical linguistic

enterprise for well over the past hundred years or so, because it acts as a stand-in for, or as a first

approximation to, a theory of language change .

The comparative method does at least part of the job of a hypothetical theory of change, but in the reverse

order. The primary role of the comparative method is in developing and testing hypotheses regarding

genetic relatedness. Its secondary, and subsequent, role (in what might be termed “realist” comparative

linguistics) is in recovering antecedent language states through reconstruction. 4

In order to demonstrate that the members of some set of distinct linguistic systems 5 are or are not

genetically related, one must demonstrate:

i that there are similarities amongst the languages compared, and then

ii that those similarities can best be explained (or can only be explained, depending on just how

confidently one wants to present the results of the method) by assuming them to reflect properties

inherited from a putative common ancestor. 6

What permits us to make the move from the observations of cross-linguistic similarity in (i) to the conclusion

(ii) that the languages in question are genetically related is an implication (rule of inference, or warrant) that

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274794

Page 2 of 23

2. On the Limits of the Comparative Method : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:28 PM

(ii) that the languages in question are genetically related is an implication (rule of inference, or warrant) that

might be stated informally as follows:

The major warrant for genetic inference

If two or more languages share a feature which is unlikely to have:

i arisen independently in each of them by nature, or

ii arisen independently in each of them by chance, or

iii diffused amongst or been borrowed between them

then this feature must have arisen only once, when the languages were one and the same. 7

A genetic argument, then, consists in the presentation of a set of similarities holding over the languages

compared, and a demonstration that these similarities are not (likely to be) the result of chance, nature, or

borrowing/diffusion. A genetic argument is thus a negative argument, or an argument by elimination, what

in classical logic is termed a disjunctive syllogism . One rules out all but one of the logically possible

accounts of relations of similarity, so that only inheritance from a putative common ancestor remains.

2.1 The first premise of the comparative method

It is not unusual for scholarly papers on historical linguistic topics, and linguistics courses on the

comparative method and its application, to deal with the possibilities of chance resemblances between

languages, and of resemblances through borrowing/diffusion. The possibility of natural resemblance is

addressed much less often. By natural resemblance I intend those instances of similarity between linguistic

objects that are simply not surprising, and do not, by their nature, call for any account. In order to be any

more precise, we must permit ourselves to be informed by insights from what can be termed “classical

semiotics,” in particular, to the semiotics of the late nineteenth-century American philosopher C. S. Peirce. 8

Peirce's semiotics involved a number of three-way distinctions – Peircean trichotomies. The best-known is

one based on a sign form's “fitness to signify”:

i indexical signs , whose forms are fit to signify by virtue of being part of their object;

ii iconic signs , fit to signify by virtue of some similarity between the sign form and its object; and

iii symbolic signs , fit to signify by virtue of some convention or agreement that their forms will stand

for particular objects.

As Saussure pointed out, only in the case of symbolic signs is the sign relation arbitrary. Since indexical and

iconic signs are natural (non-arbitrary), we have no reason to be surprised by their cross-linguistic

similarity. It is only in the case of arbitrary relations between the form and the meaning of linguistic signs

that comparativists ought to find cross-linguistic similarity surprising. Comparative historical linguists only

have cause to be surprised by, and must seek explanation for, similarities between form-meaning pairings in

different languages when those pairings are symbolic .

So the comparativist is on the safest ground by restricting comparison to those linguistic signs that are the

most arbitrary and conventional – individual lexical items. One has no strong warrant to infer genetic

relatedness from similarities in iconic signs – onomatopoeic forms, metaphors, compounds, or syntactic

patterns – since such similarities can be explained in terms of the limited possibilities afforded by

observation and analysis of the world. 9 I will refer to the restriction of comparison to symbolic signs as the

semiotic restriction on, or the first premise of, the comparative method.

It is, therefore, the first premise of the comparative method that focuses attention on the lexica of the

languages compared, and not the fact that nineteenth-century linguists couldn't do syntax, or anything of

the sort. At the risk of unnecessary repetition, we have no clear warrant to compare anything other than

symbolic linguistic signs, because sign similarity is only surprising when the signs are symbols. This fact

does not mean that we must restrict comparison to monomorphemic signs, but it does mean that we are on

increasingly thinner comparative ice the more abstract/less symbolic the signs we compare.

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274794

Page 3 of 23

2. On the Limits of the Comparative Method : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:28 PM

2.2 The relation cognate with

It is tempting to think of the relation cognate with as differing only in domain from the relation genetically

related . The latter, defined over languages, would be in some sense the sum of instances of the relation

cognate with , defined over individual linguistic expressions, grammatical rules, or whatever. But that

interpretation confuses reality, what actually is the case, with demonstrability, what we can show to be the

case on the basis of available evidence and “the state of the art.” Two languages 10 can, in principle, be

genetically related without a single cognacy relation being evident in the synchronic states of those

languages. That is, those languages might be genetically related, without our being able to adduce any

evidence of that relatedness. And that is precisely what instances of the cognate with relation are – a

demonstration of genetic relatedness. If one can prove that even one single cognate pair holds over two

languages, one has proven those languages genetically related. 11

Two linguistic objects σ 1 and σ 2 are cognate:

cognate(σ 1 , σ 2 ) [≡ cognate (σ 2 , σ 1 )]

iff both are reflexes of a single antecedent linguistic object *σ:

reflex(σ 1 , *σ 1 ) 㱸 reflex(σ 2 , *σ 2 ) 㱸 *σ 1 = *σ 2

A linguistic object σ t is a reflex of 12 a linguistic object σ t , if:

i σ t and σ t , are in temporally distinct language states t and t' (t subsequent to t’) and if:

ii σ t is a “normal historical continuation” of σ t

Being more precise about what is meant by “normal historical continuation” isn't easy. It must involve notions

like “normal language acquisition” and “normal language change.” 13 Although there may be some danger of

circularity here, it seems to me safe to assume that historical linguists will know what I have in mind.

As noted above, comparative historical linguists must identify instances of the cognate with relation in order

to demonstrate genetic relatedness. Even the techniques of “mass comparison” (as evidenced, for example, in

Greenberg 1987; Ruhlen 1994), or any other method that begins with the mere observation of similarity,

must ultimately trade in cognates. There is no other logical possibility, in the absence of written records or

time machines. The comparative method is simply the principal (indeed, the only) means available to

historical linguists for identifying cognates convincingly .

2.3 Phonological comparison and the regularity assumption

Let me stress this point again. The relation cognate with is independent of the comparative method. Though

the comparative method is a technique for identifying cognates, cognacy can exist without the comparative

method being able to demonstrate it. That is, the comparative method has limits.

The most immediate limit on the method is the one faced by the working comparative historical linguist

even before she or he sets off to hunt for cognates. The problem is where in language to look for cognates.

One could look anywhere (a point taken up below with regard to grammatical comparison in section 3.1). But

the comparative method, I would argue, is not designed to demonstrate cognacy in general, but cognacy

only in the lexicophonological domain .

For the remainder of this section, I will assume that candidates for cognacy testable by the comparative

method are (possibly morphologically complex) linguistic signs whose phonological shape is in a form no

more abstract than (taxonomic) phonemic. That is, I assume we are comparing morphemes or morpheme

sequences, in phonemic notation, up to the level of the phonological word.

As observed at the end of the preceding section, the comparative method is a procedure for identifying n -

tuples that are instances of the cognate with relation, at some reasonable level of confidence. I will assume

that any pair of items f and g , from different languages and meeting the domain conditions, are potential

cognates. And I will use the possibility operator M of modal logic to represent potentiality. The problem of

proving cognacy for potentially cognate pairs can be reduced to or recast as the problem of defining a rule

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274794

Page 4 of 23

2. On the Limits of the Comparative Method : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:28 PM

proving cognacy for potentially cognate pairs can be reduced to or recast as the problem of defining a rule

of M-elimination that licenses the move:

M cognate( f, g )

cognate( f, g )

The comparative method is an attempt at defining this rule of M -elimination. The following is an informal

approximation:

M-elimination

A pair ( f, g ) of potential cognates is a cognate pair if:

i they meet a similarity condition: that f and g are similar in both facets of the sign relation, in form

and in interpretation, and

ii they meet a disjunctive elimination condition that the similarity is not (likely to be) a consequence

of chance or of borrowing/diffusion .

2.3.1 The similarity condition

Condition (i), the similarity condition on potential cognates, is logically prior to condition (ii), on non-genetic

accounts of the similarity. After all, you have to recognize similarity before you seek to explain it! But that

fact does not make the similarity condition a precondition (that is, a condition on potential cognacy), as

often seems to be assumed. I choose to view condition (i) as part of the proof of cognacy (as part of M -

elimination) because I believe that the definition of similarity is in fact part of the comparative method, at

the very least, as the method was first devised.

Under this interpretation, it is the similarity condition of the comparative method that rules out natural (i.e.,

iconic) similarities and enforces the semiotic restriction on the comparative method. With the comparative

method, we restrict comparison to symbols because it is only similarity between arbitrary and conventional

( symbolic ) signs that is surprising, and that could be evidence of cognacy.

Similar symbols must be similar in both form and interpretation. While it may not be entirely fair to say that

comparativists have done nothing to clarify the notion “similar meanings,” we haven't done much. Most

recent work has focused on grammaticalization , 14 the process by which reference to particular sets or

relations in the world changes into higher-order reference: motion verbs to source/goal markers, object-part

relations (like “top surface” or “cavity beneath”) to object-location relations (like “on” or “under”), and so

forth. But we are still very much at the data-collection stage in this endeavor, and are informed in it only by

vague senses of what are possible metaphors or metonymies. Sadly, we don't really pay much attention to

the meaning side of things. In general, unless a particular meaning comparison grossly offends some very

general sense of metaphor, it' s “anything goes” with regard to meaning.

Comparative historical linguists have been rather more careful in stipulating what it means for linguistic

symbols to be similar in form. Observe first that similarity of form must be complete similarity. Put rather

brutally, if the front halves of two forms are similar, but the back halves aren't, then the forms are not

similar. In practice, we observe this condition by segmenting each form into its component (segmental or

autosegmental) parts, and then mapping the segmented forms into a set of correspondences between a part

or parts of one form and a part or parts (possibly nil) of the other. We need not go into the mechanics of

that segmentation process here. The problem of the similarity of sign forms then reduces to the problem of

similarity of objects in a correspondence relation. And that, as we shall soon see, is not a problem at all!

Feature (attribute-value) theories of phonological representation (and of articulatory description that

precedes them) make it possible for us to measure the similarity between two representations of

phonological form, in terms of shared attribute-value pairs. Phonological feature theories do not, of course,

tell us precisely how many attribute-value pairs must be shared by two forms for them to be deemed

sufficiently similar to be cognate. Nor is it clear how one would, in practice, begin to construct a method

that makes such a determination.

2.3.2 Regularity, similarity, chance, and borrowing

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274794

Page 5 of 23

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: