Commentary on Crowley, Ch. 8-12

Commentary on Crowley, Ch. 8-13
Subgrouping, Wave Theory,
Language Contact, Areal
 Shane’s presentation raised a number of practical questions
such as:
 Which dialect is the most conservative and which is the most
 According to Shane’s lexico-statistical analysis, Matu-Daro
retains the most cognates, Kanowit the least. (I have
suggested that he re-do his percentages based on just the first
200 words of our list, to avoid the problem of artificial gaps
in the Kanowit data.)
Innovations imply out-migration
 What does Shane’s classification imply with respect to the
movement of the Melanau people? Did they start in Kanowit
and migrate to the coast? Or did they come in from the coast
and migrate to Kanowit?
 According to two articles by Blust (1991) and Ross (1991),
conservative dialects typically represent stay-at-homes,
whereas innovating dialects imply populations on the move.
Why should this be?
 Ross (1991) offers a sociological explanation relating to the
“correction behavior” of migrating vs. sedentary adults.
 The innovators are the children acquiring the language.
Correction behavior of adults
 To the extent that the children are corrected,
innovation is curtailed; to the extent they are not
corrected, innovations develop rapidly.
 According to Ross (1991), migrating populations are
less concerned about the niceties of pronunciation and
grammar than are sedentary populations.
Preliminary Method of Subgrouping
 Lexicostatistical analysis is useful as a preliminary tool in
 Remember our results obtained after only an hour or so of
such analysis and reported in Homework #1.
 Based on the first 200 words of our list, we discovered that
the pair [Matu-Daro, Belawai] shared the highest number of
cognates; next were [M-D, Dalat]; and last were [Dalat,
Advanced Method of Subgrouping
Lexicostatistical methods seek to establish subgroups by
counting words/cognates.
The Comparative Method achieves the same goal by
counting shared innovations (rules, including sound
changes,and morphological (analogical) changes).
Where the results differ, you can have an argument about
which method represents “the truth”.
For 99 out of 100 linguists, the answer is clear: the
Comparative Method is the only reliable approach.
Although I agree with the 99%, I do not
share the attitude of some librarybound linguists that lexicostatistical
analysis has no value.
Therefore, I find it interesting when the results of the two
methods converge, and I am also interested in knowing why
in case the results differ.
So what about Melanau?
 Null hypothesis: all four dialects are sisters.
 Last resort: The null hypothes is maintained
until the evidence forces a better one.
 Disinterestedness: A scientist shouldn't be
invested in the outcome. Hypotheses are not
“good” or “bad” but only supportable or
unsupportable based on the evidence.
So what about Melanau?
 To look for a subgrouping hypothesis, we need to count not just
cognates, but also (and especially) rules.
 Any two rules shared by any two dialects is POTENTIALLY a
shared innovation. (Remember this term.)
 In the Comparative Method, a subgroup is defined over the number
and quality of shared innovations.
 For example, one group of German dialects share the First
Consonant Shift (Grimm’s Law), e.g. PIE *p>f) and the High
German Second Consonant Shift, e.g. PGmc *p>pf).
Great Vowel Shift = 10 or more sound
English dialects show evidence of the GVS. A language closely
related to English that failed to undergo the Great Vowel Shift is
barred from membership in the same lower order subgroup.
Such a language is Frisian.
Of course, English and Frisian do belong in a higher-order
subgroup that also includes Dutch.
No end in sight
Subgrouping goes on and on. PIE has three
primary “branches” representing first-order
subgroups: Anatolian, Hellenic, and IndoIranian. Each of these has three or more
branches; and each of these has more and
more branches, all the way down to hundreds
of individual languages.
PAn and PMP
Proto-Austronesian has nine primary branches, all
representing languages currently spoken on the island of
One of the nine, Proto-East Formosan, is the mother of
Tree theory vs. Wave theory, p. 249
 An interesting conumdrum exists at the heart of the field of
Historical and Comparative Linguistics.
 There exist two perfectly valid theories of the way “languages
change and people move”.
 One is called the “tree theory” and/or “theory of divergence,” and
it has served as the basis of this course.
 It’s not unlike the decision to teach articulatory phonetics. There
exists another approach (acoustic phonetics), but it would be
confusing to teach both at the same time.
Early European Migrations
Branches imply a tree; waves imply a
pond (or a flat map)
 Tree theory assumes that people move and languages change
without looking back. Here today, gone tomorrow. It
follows that languages will keep “branching” as they lose
contact with their ancestral roots. Tree theory works best
over ever larger tracts of time. PIE and PAn go back 6,000
years. That’s plenty of time for languages to diverge in treelike fashion.
 But dialectologists have long known it’s not like that “on the
ground”. Changes actually begin in one location and “spread”
to the next location, like ripples on a pond.
Wave theory
 Moreover, speech communities are not monads; they interact
with other communities. It’s the same with rules (changes).
 Often many changes (rules) will spread from one language to
another within an area. Crowley mentions the spread of uvular
[ʁ] in Europe (p. 260) and the Rhenish Fan (p. 247) as examples.
 Accordingly, linguistics has developed a sub-field called areal
 A principal tool of Wave Theory is the isogloss.
 The term is derived from the Greek and means “same word”.
 An isogloss is a line—often in the form of a closed circle like
a ripple on a pond—showing the spread of a new word or a
sound change over a linguistic area.
 A “bundle” of isoglosses defines a dialect area.
“Isoglosses” vs. “shared innovations”
 One term is tree-theoretical, the other is wave-theoretical.
 Both are used to define dialect groupings.
 Both are statistics-bound in the sense that dialect groupings
depend on convergence of significant numbers of valid
 For example, in wave theory dialect boundaries are defined
in terms of “bundles” of isoglosses.
Wave-like change across languages
A famous example concerns retroflection in the languages of
India. There are at least 14 major language groups in India;
nearly all of them have retroflex alveolar stops. For most of
these languages, retroflection is “irregular”—i.e. does not
follow from the respective protolanguages. Probably it was
borrowed in ancient times, probably from Sanskrit, and
spread across the map.
Wave-like changes
 Another example is the change *q > h/__# in Malay and scores
of other languages in Southeast Asia.
 This word-final –h corresponding to PMP *q is irregular in many
of the languages, although it is regular in Malay and Javanese.
 Malay and Javanese were the prestige languages and power centers
during the reign of the Sriwijaya Empire (7th-13th century). It is
necessary to assume that the RULE was borrowed by scores of
languages, which accounts for the unexpected appearance of
word-final –h in languages scattered all over the map.
Rejang irregular -h
In Rejang, there are three dialects in contact with Malay, and two dialects insulated
geographically from such contact.
Two of those in contact with Malay have developed word-final –h reflecting PMP *-q; the
other three reflect PMP final *-q as glottal stop.
The irregularity arises when one considers the structure of Proto-Rejang. The ProtoRejang reflex was clearly glottal stop (simplicity, Uniformitarianism, etc.). So how did
two dialects develop –h? There is no evidence of *Ɂ > h except word-finally in the two
dialects in contact with Malay. And in one of the dialects—Rawas—this –h is the only
allophone (Rawas lacks word-initial and intervocalic /h/).
Wave theory to the rescue
 A convenient conclusion is that Rawas and
Kebanagung dialects borrowed -h from Malay as a
consequence of intermarriage with Malay speakers.
 Bilingualism and the prestige of the loaner language
are two material causes of most borrowing, which
in turn has an effect on the children. (Always the
children are the primary agents of change.)
What to do about a conundrum?
Given that we have two theories—tree
theory and wave theory—which one is
Duh! They are both correct!
 One works best over long stretches of time and among
languages that have lost contact with one another.
 The other works best “on the ground” over short stretches of
time, and among languages (and dialects) in constant contact
with one another.
When are both relevant?
 Theoretically, both are always relevant all the time. But it’s
confusing to mix them indiscriminately.
 In practical terms, when problems arise within a tree-theoretical
analysis, a solution can sometimes be found by adopting a wavetheory approach.
 This is especially true when dealing with closely-related dialects,
such as Melanau—and especially Matu and Daro (which are taken
to be a single dialect); and Matu-Daro and Belawai. The closer the
dialects are geographically, the more likely they are to borrow not
just words, but rules.
Areal Linguistics, p. 261
 Moreover, rules may be borrowed across a wide area from a
language, such as Malay, which is (or was once) held in high
 Such was likely the case with the Rejang final -h in two of five
dialects, and which upset the neat tree-theoretical applecart.
Wave theory came to the rescue to explain the odd -h.
 Just as borrowed words can be set aside when establishing
proto-languages within tree-theoretical assumptions, so
borrowed rules can and must be set aside, i.e. dealt with
Substrate theory, p. 270
 A linguistic substrate refers to indigenous languages that may
have become extinct as the result of contact with, and
colonialization by, a superior culture.
 Substrate languages often leave traces in the form of
vocabulary items and even rules.
 For example, the NYC bowery ‘dis’ and dat’ may be
remnants from a Dutch substrate (NYC < New Amsterdam).
 Psycholinguistic test: Name five American rivers.
Again, we set aside substrate
influences. Why?
 Because reconstruction is tree-theoretical; borrowing
patterns have their place in wave-theory approaches.
 A proto-language belongs to the theory of divergence.
Remember the second part: “Barriers reduce the density of
 By contrast, when “people move” next door to a robust
community speaking a different language, both languages
may change in part because of the contact. Comparative
reconstruction breaks down. Wave theory rules here.
Does tree-theory apply to Melanau?
Clearly the answer is yes. The dialects
are far enough apart linguistically (and
geographically) to warrant applying the
Comparative Method.
Can our four dialects be subgrouped?
 That remains to be seen.
 It’s no loss if they can’t be; we just want the facts.
 Two tree-theoretical possibilities remain on the table.
All four dialects are sisters: W+X+Y+Z (null hypothesis)
Two or three can be subgrouped: W + XYZ or WX+YZ.
Back to the question of possible
shared innovations
 Take the rule *-k > -Ɂ in Dalat and Kanowit.
 Is this a shared innovation?
 Remember I said: Any two rules shared by any two
dialects is POTENTIALLY a shared innovation.
 Just looking at *-k > -Ɂ in isolation, it is impossible to tell whether it is a
shared innovation.
Candidates for shared innovation
status; problems and solutions
 Part of the problem with*-k > -Ɂ is that it is a common (natural) change
that might have occurred independently in each dialect.
 Moreover, the same change has affected Malay.
 What is needed is more evidence that the two dialects in question indeed
form a subgroup.
Shared innovation is a technical term.
 In order to count as a shared innovation, a rule must be established to have
arisen within a common subgroup.
 No single rule can have such status—with one possible exception (see
below). What is needed is a goodly number of identical rules converging on
a set of dialects.
Strong and weak evidence
 Some candidates for shared innovations have more weight than others as
evidence for subgrouping.
 Shared uncommon rules (such as Dalat *a>i) have more weight than shared
common ones (such as *-k>Ɂ).
 Weirdness has its uses in Historical Phonology.
More research needed!
 The way to strengthen a weak subgrouping
argument is to find support in the form of other
 Are there any other candidates?
 I wish I knew!
 That’s what we have graduate students for!
Once Over Lightly
 The last slide lists a set of topics I found interesting when
reading the last five chapters of Crowley.
 Please refer to the Study Guide for a couple of possible test
questions designed to help focus your reading.
Once Over Lightly
 Neogrammarian Hypothesis, p. 226
 Cultural Reconstruction, Chapter 13
 Blust’s attempt to reconstruct “iron” for PMP, p. 316
 Age-area Hypothesis, p. 305
 “Paleo-linguistics”, p. 308
Commentary on Ch. 8-12
LING 485/585
Winter 2009

similar documents