<Language> Lass, part 1 of 3: Electronic Virtual Review

[H. Mark Hubey]

<><><><><><><><><><><><>--This is the Language List--<><><><><><><><><><><><><>

This was meant originally to be a review of the book for the
Linguist, but when I finished typing in the parts I wanted to
make comments on, it was already 80 KB. It was too long and
nothing that long could be published in the Linguist.

I probably will be tied up with other things this summer and
beyond, so I decided instead to 'release' this to <Language>.

I would like to make comments on these later, but everybody
else is welcome to do so.

The material in braces {..} is probably all mine and was meant
to leave markers for myself.

----------------------Part 1 of 3 ------------------------

Lass, Historical Linguistics, 1998.

p. 25
The cornerstone of historical rationality is this simple proposition:
there are no miracles. This reflects the epistemological stance
generally called  uniformitarianism: the principles governing the world
(=the domain of enquiry) were the same in the past as they are now.

 	General Uniformity Principle

	Nothing that is now impossible in principle was ever the case in the
past...The principle can (with some risk and a lot of tact) be taken
further. Not  only did nothing impossible ever happen, but in general
the most likely things have always been the most likely to happen. There
is a second or derivative Uniformity Principle, which is statistical or
probabilistic:

	Principle of Uniform Probabilities
	The general distribution of likelihood in a given domain was always the
same in the past as it is now.


Again, this must not be taken simplistically. The likelihood of a
resident of New York being a Yiddish-speaker in 1800 was not the same as
it is now, that of a Londoner having a telephone was zero in 1856, but
is quite high now, and so on.

p.28
	General Uniformity Principle
	No linguistic state of affairs (structure, inventory, process, etc) can
have been the case only in the past.

	Uniform Probabilities Principle
	The (global, cross-linguistic) likelihood of any linguistic state of
affairs (structure, inventory, process, etc) has always been roughly
the same as it is now. ... Now one possible objection to this kind of
uniformity argument (cf. Matthews  1981, Lass 1986a:233ff) must be taken
account of. It goes like this: how good really is `the best of our
knowledge'? Surely nobody has examined all the languages spoken in the
world at present (not to mention all past ones, including those that
vanished without a trace, and all future ones).

	{This is where the model is really needed. The probabilities and
	etc come from the model. i.e. low number of phonemes in the past,
	etc, consonant-clusters in Europe etc}

Therefore the argument is not based on knowledge at all, but on
ignorance; it's only failure to recognize the obvious (even necessary)
limitations of our knowledge of the `total set of human languages' that
allows this smugness.

The counterargument is essentially philosophical. All human knowledge is
flawed, provisional and corrigible, which is what makes scholarship of
any kind worth doing.

p. 104		{Relatedness, ancestry and comparison}

p. 105
Any one could be true for a given case: there are coincidences, like
Modern Greek mati and Malay mata `eye'...or Rumanian and Lau
(Austronesian) dori `to desire'.

p.106
Chance (a last, not first resort if we're looking for an interesting
architecture) is virtually ruled out by multiple sets with strong
internal resemblance, but none to other such sets....We are left with
common ancestry. As a first approximation, a family is a set of
languages with enough marked and pervasive similarities for us to want
to consider them to be `genetically related'. That is, a language family
is a lineage, which means there was a  common parent. Oversimply, they
were once the same language.

The idea of linguistic `descent', i.e. monogenesis and subsequent
differentiation is an ancient one; the locus classicus in the
Judeo-Christian tradition is the Babel story.

p.108

The notion of some kind of relationship between languages other than
monogenesis, specifically the possibility of multiple (unrelated)
lineages seems to have taken root after the introduction into the west
of Hebrew grammatical studies in the sixteenth century (Percival
1987)....

We ought to note also that the idea of relatedness [of Conrad Gesner]
was based on essentially (as with Jones) on phenotypic resemblances,
even cultural ones. One of the triumphs of later comparative method was
the transcendence of a superficial (phenotypic) notion of `similarity',
and the development of a quite different methodology, based to be sure
on `shared properties', but not on obvious similarities. It was this
(largely but not exclusively: there are clear  adumbrations in Sajnovics
for instance) nineteenth-century innovation that made truly genetic
approach to language filiation possible.

p. 109
...while metaphors of `parenthood', lineal relationship, etc are useful
and indeed necessary in historical linguistic discourse, the unmarked
form of linguistic parentage is parthenogenetic.

We assign languages to a family first because they (intuitively) have so
many have consistent similarities that we reject chance or borrowing as
the causes; but we also recognize that the family members are still
sufficiently distinct to be named as individuals. Different sets
establish different language families. Historically, similarities of the
requisite type suggest common ancestry; differences within
similarity-sets suggest, in Darwin's famous phrase, `descent with
modification'.

p.110
Evolving systems (probably in any empirical domain) are characterized by
what is known in mathematical theory of (nonlinear) dynamical systems
(popularly `chaos theory') as `sensitive dependence on initial
conditions'. The smallest differences in initial conditions may, in the
later development of any pair of sublineages starting from (essentially)
the same point have enormous consequences. In biotic and cultural
systems, this connects with a property called path-dependency: `isolated
populations... must tend to diverge even when they start from the same
initial condition and evolve in similar environments' (Boyd & Richerson
1992:186). The key here is `similar': given any initial isolation
whatever, and sufficiently complex and hence contingentlcy-laden
landscape over which the original split populations develop, no two
evolutionary trajectories (as wholes) will ever be identical. Since the
very fact of initial `isolation' (whatever causes  it) means that the
initial conditions will always be slightly diffferent, linguistic
evolution will by definition be subject to massive contingency,
which guarantees its genuine historicity.

{He's adding this bit about isolation etc. the connection to nonlinear
DEs is neither immediate nor obvious. He does not explain how these
systems can be represented as trajectories}

{we can consider members of a language family simply as sample functions
of a stochastic process in N dimensions! Why do we have to give each
sample function a name ?}

Historicity then implies some kind of `irreversibility'.

{No, it does not. He makes this up. Explain the Einstein Brownian motion
problem and Fokker-Planck, Kolmogorov, and entropy.}

On the macro-scale this is indeed the case for linguistic evolution:
there are frequent `repetetions' of sorts but (except at a certain
micro-scale..) they are never total.....
{Are there only two scales? what are they based on?}


Historically evolved entities at a certain scale (see below) are
contingently individual, hence not repeatable; the topography of the
epigenetic landscapes over which they emerge and d___lop is too complex
and locally differentiated to allow exact repetition. The arrow of
evolutionary time (biological, linguistic, cultural) is not of course
thermodynamic (entropy-increasing); but it is qualitative.

{Hard to understand what he means!}

And historicity itself acts as its own guarantee of both continuity and
ever-increasing differentiation, because any given state is the product
of all prior one.... Irreversibility and increasing differentiation (in
short historical `uniqueness') then may be functions of the complexity
of  initial conditions or system structure or both.

This criterion seems to fail at one linguistic micro-level: in
phonological change. Here the vocabulary of elements is (comparatively)
small, and what look like exact repetitions are indeed not uncommon. We
often seem to get evolutionary reversals or retrograde changes, where
some phonological category X goes to Y (and to Z...) and then X or Y is
generated again.

p.113
In biosystematics of the type called cladistic (<Gr klad `twig') which
will be a roudh model and source of terminology here, clades (taxa) are
established solely on the basis of innovative or derived characters
(apomorphies); shared innovations (synapomorphies) define subtaxa. A
cladogram (tree) for any cluster of lineages is `a hierarchical scheme
of nested synapomorphies in which the synapomorphies define classes or
taxa ' (Crane & Hill...) In this light, retained ancient or primitive
characters (plesiomorphies) are not per se of particular interest at any
given level; though shared retained characters (symplosiomorphies) may
be at higher levels. (Since the concept of cladistic relationship is
hierarchical, different characters will be significant at different
levels of `universality' or inclusiveness: Crane & Hill, 140)...

{Here are two problems already; It has all the weaknesses of biology
which is also rather new on the quantitative science list. There is no
mention of distance which biology can also use. Secondly, the concept of
primitive  characters or features is not yet in linguistics.}


Since cladistic analysis depends on the distribution of characters, and
in particular their identification as innovations or retentions, we need
criteria for making such decisions. It is possible in principle to
construct biological cladograms with no historical information (i.e.
without recourse to  paleontology or stratigraphy: Crane & Hill
1987:144). But this is not normal practice; the investigator usually
`polarizes' characters, on the basis of an `ingroup' (the putative clad
under investigation) and an `outgroup' (some other putatively related
clade or cluster of clades).

p.115
But most fields show general agreement on what (in a given character
type) should count as primitive, and what as apomorphous. Such
identification is crucial, since we want phylogenies, not typologies.

{Still hasn't grasped the significance of what is being said.}

p.118

Establishing phylogenies would be relatively simple if the evidence
always led to a set of infinitely diverging taxa, with all characters
unambiguously primitive or derived. But both in biology and language
there are events that lead to merging of paths, creating reticulate
rather than continuously branching graphs; or to the occurrence of the
same (primitive or derived) character on different branches, at the same
or different hierarchical levels....

The phenomena producing such problematic graphs are generally grouped
under the term homoplasy (Stevens 1987). The main types are:

1. Convergence. Two or more lineages develop the same character
independently
2. Hybredization or transfer. A char is transferred from one lineage to
another
3. Reversal or retrograde change. A character re-appears after it has
been lost, i.e. an archaic or primitive state appears in a branch
containing its own apomorph at a higher level.


p. 130		[back to languages]

`Family resemblance' is no longer the simple `phenotypic' notion
suggested by the first examples and by Jones' comments; it is now both
formal and `genotypic' (in the loose sense that lineage, not appearance,
underwrites the relationship). It is not even necessary for there to be
any phenotypic resemblance at all between two languages for them to be
assigned to the same family, as long as transformations over time can be
formulated rigorously.


{He is now falling into the myopic vision. If that were the case we
humans still would be called fish. It is only common sense that puts us
in the same category as the apes, and closer to dogs than to fishes. But
in linguistics this has been done away by making up words for concepts
that are lacking.}

Therefore:

1. A (perceptible) resemblance counts for nothing per se(e.g architect
Fi arkkitehti)  though it may be a useful heuristic because it leads to
falsifiable results

2. A resemblance counts only if it can be shown to be `regular'

3. A  non-resemblance counts if it can be regularly derived from an
ancestral form

4. Regular correspondences are assumed to have a genetic origin:
regularity = (ultimate) monogenesis

....
We can now define comparison and correspondence more precisely:

1. Two formatives (words, morphemes, etc) fi, fj in two languages Li and
Lj may be compared in the first instance iff they show enough semantic
similarity to suggest a priori that they have a common origin. (Or if
there is already evidence for a corresondence in which they might
participate, and the semantics can be made to fit without excessive
ingenuity.)

2. A phonological segment si in fi and a segment sj in fj correspond iff
the relation between them can be shown to hold constant for a  majority
of items in both languages in which they appear in teh same context.

3. A likeness or even identity of a given segment in fi with the one in
fj is not a correspondence, even in the presence of semantic likeness or
identity, unless it is `regular' in the sense of 2.

But two caveats are necessary. The first is suggested by the discussion
of homoplasy..

4. If a character is one of a small possible subset that tends to
repeate anyhow, its taxonomic significance is at least suspect.

This distinguishes typology from phylogeny: e.g. possession of an aspect
system or /f/ in itself is unlikely to be interesting, because languages
either have them or don't, more or less at random; but the particular
morphs constituting an aspect system, or the lexical set in which /f/
appears, can be taxon markers, if their material basis can be shown to
derive from regular correspondences.

{This is nonsense. we only know carbon-based life forms and carbon
based intelligence on this planet but certainly if and when we have
AI, [and it is silicon based], would we claim that the carbon-based
ones are not genetically related to each other?  Of course, if we
took what are merely called typology and made them primitives we would
then count that as the most important aspect of genetic relationship.
Everything is still circular. This also cannot distinguish borrowing
from "geneticity". what good is a method that only works when we know
almost everything about a language family? It should work for those
about which we know little. One can try Bayesian logic here but the fact
that there might be determinism means that we need to first assign the
probabilities, but to do that we need to first obtain the probability
density
that includes the deterministic aspects of language change, and that has
to
be done via Fokker-Planck methods, and that implies that we have to know
the
deterministic equations of evolution or make assumptions for them.}


The second derives from certain kinds of known contingencies in language
history;

5. If a language has been in intimate contact with a cognate language
with a different history, extensive borrowing may produce multiple
correspondences, which have to be sorted out on other grounds.


...Finally, given the above, similarities in `overall type' are little
or no value in filiation: the fact that a group of languages shares a
particular kind of syntax (ergative, nominative-accusative), morphology
(inflecting, agglutinative) or phonology (non-pulmonic airstreams, tone)
may be of areal or sociolinguistic or historical  interest, but cannot
in general be taken as genetically indicative.

{He mixes up different scales. Are we not all descended from fish? back
and forth}

p. 140 Variation, diffusion  and competetion

Aside from the interfering effects of analogy and borrowing, which were
incorporated in the `classical' nineteenth-century procedure, there are
other difficulties. We know now, due to the painstaking work done in
the  last three decades in both the Labovian `variationist' and `lexical
diffusionist' traditions (Chen 1972 and countless other contributions),
that the following propositions are generally true for internal
(`evolutive') change:

a. Change is not catastrophic, but takes time. {Often centuries}

b. The time is occupied by a process of variation: primitive and
apomorphous states coexist, with cumulative weighting heading towards
categorical implementation of the latter.

c. Changes may never complete, but may abort at virtually any stage.The
typical shape of a change in progress is an exponential curve heading
for an asymptote.

d. Changes may compete for the same environment, and will operate under
conditions (b,c) during their competetion (WAng 1969)

e. Changes may recede over time (lexical fading), according to the
conditions (b,c) giving results that may look like gradual
implementation rather than loss.

f. Changes filter or diffuse through available environments rather than
hitting all at once. In cound change these environments are typically
item-specific (either lexical or morphological), so that at a given
stage only certain lexical items or morph(ems)s will be affected.

g. Changes are subjected to constraints and conditioning based on social
factors like age, gender, style, class, etc.

p. 160

This does not however compromise the freedom of `lumpers' to claim
superiority to `splitters', and posit increasingly inclusive hypertaxa,
from the relatively modest (`Indo-Uralic', Elamo-Dravidian) to the
monstrous (Nostratic, Eurasiatic, Amerind). These enormous taxa ...

One of the oldest and perhaps the most famous of these hypertaxa is
`Nostratic' (or its extended avatar, `Eurasiatic). This mildly
ethnocentric term (Our FAmily, comes from Holger Pederson 1924) but the
basic idea was already broached in a semi-popular form by Henry Sweet in
1900. Sweet (1900: ch 7) claims that Indo-European (Aryan) has `strong
affinities' implying ultimate monogenesis, with at least two other
families and one apparent isolate: `Ugrian' (Finno-ugric,and especially
Balto-Finnic), `Altaic' (Turkic, Manchu-Tungus, Japanese) and Sumerian.
According to Sweet (123) `the Aryan languages are a branch of the great
Ugro-Altaic family...extending now from the Pacific to the Atlantic with
hardly a break'. Sweet is (except for Sumerian) less inclusive than
Pederson; and `Nostratic', has grown and changed over the years in many
ways, and has been joined in the field by other hypertaxa like
Greenberg's `Amerind' (1987) which includes most of the languages of the
Americas.
...
Some sort of Europe-Asia connection has been mooted for a long time; one
of the most striking features suggesting it is the `Mitian' (or
`me-thee' " Nichols 1992) phenomenon: the presence of a formative
containing /m/ as a marker for the first person (pronominal and/or
verbal) and one with /t/ for the second person....

Even `the most superficial comparison' of Finnish with Indo-European
`would be enough to establish a common origin' (Sweet 1900:116)....

Now it would be difficult enough to imagine (especially where the
languages involved are known to have been geographically contiguous for
a long time, and where there has been extensive loan-contact) that
similarities like these do not involve some kind of `genetic affinity'.
But of what kind?

..the issue is a live one now, because so much of the lumping tradition,
whether Nostratic or other, bases its claims  on criteria that are at
best different from, at worst looser and sloppier than, the traditional
ones. And this is aggravated by a somewhat  disingenuously Pickwickian
use of the terms `etymology' and `cognate', to the extent that one man's
etymology is another man's word-list.

... I will show below that Ruhlen's and others' claims of this kind are
based on a tendentious misconstrual of a typical etymological dictionary
entry.

According to Ruhlen (and his mentor Greenberg)....essentially says that
valid filiations can be established by `multilateral comparison'. That
is, simpy collecting vast numbers of lexical items that look too similar
to be coincidences, and then calling such collections `etymologies', and
their contents `lexical cognates'.

p.163
The problem with such procedures (and they appear to be worse in the
American Indian linguistics...) is nicely summed up in Lyle CAmpbell's
critique of Greenberg:

What Greenberg calls etymologies are actually sets of word equations...

The Greenberg approach is retrogressive. All `similarity'-based (purely
phonetic) taxonomies, if they claim to be phylogenetic,  miss the real
point: phylogenetic taxonomies are based on  homologies not analogies.
And in linguistics the only solid defn of homology is `regular
correspondence', which is itself justified only by the possibility of
reconstruction.

p.167
Are we dealing with genuine lineages, or rather with Sprachbunde or some
other kind of areal phenomenon, periods of `hybridization' and massive
diffusion? OR as Andronov (1968:32, on the Dravidian/Uralic connection)
has suggested, `newly forming languages' using as  `building material
for their structure the linguistic substance of preceding unrelated
languages or language groups?

p.168
In the end, however, the most telling argument against the rackety kind
of impressionistic inspection the Greenbergians do is that it is lousy
historiographic method. It will, in its characteristic uncontrolled use,
pick up demonstrable non-cognates that are very `similar', miss clear
cognates that are not, and along with this miss a lot of good stories
and connections.  {examples given}

p.169
The best evidence for relationships is regular phonological
correspondence; this is also what makes detailed reconstruction
possible. But there may be strong indexes of filiation in other domains,
particularly in  morphology.

p. 172			Convergence and Contact

p.186
Nobody has ever succeeded, as far as I know, in establishing any
absolute constraints on what is borrowable, though there are discussions
that would seem to make certain kinds so rare that we would not expect
them, and hence could use this rarity as a guide in reconstruction.

{Exceptions are noted and his belief is that "this common belief is a
poor historiographical guide: exceptions are so frequent that merely
invoking vague notions like `prestige' is hardly satisfactory"}

p.187
To take one more easy example, Finnish and other Balto-Finnic languages
have borrowed extensive core lexis including body-parts, kinship terms
and numerals, from Indo-European:....

{Now he is confused. According to original rules that would make Finnic
IE but now since the rules have been changed, he changes to allow
borrowing. But now he is trapped. If they can be borrowed, why couldn't
the IE family be due to convergence? Now we are back to square zero or
the Swadesh-like lists.}

p.188
Anttila in fact virtually admits this when he says `It would be
ridiculous to maintain that Finnic speakers did not have native,
inherited names for...body parts. If "prestige" itself is not the right
term, then the examples indicate very close cultural contacts and
intermarriage.'

p.194
But the unmarked case is that a foreign phone will be replaced by a
native one, and a phonotactically admissible sequence will be replaced
in the same way (hence Fi pelto `field' < Gmc */feldh-u/, Risto
`Christopher', ranta `bank, shore' < Gmc */strand/, kinkku `ham' < NGmc
skinka, etc.  Finnic historically has no voiced stops or /f/ and does
not allow  initial consonant clusters).

p. 207   	Endogeny vs contact as a methodological issue

The whole discussion resolves to one simple question: if indeed anything
can be borrowed (though not all borrowings are equally likely) how do
you tell in any given case whether borrowing has occurred? Such
judgements may (as the preceding discussion suggests) be extremely hard
to make, and very much contingent of survivals and other kinds of
evidence. In all such cases at least three distint questions have to be
taken into account initially:

1. What is there about the character in question that makes one suppose
it might be borrowed in the first place?

2. Was there a historical situation such that borrowing would be likely?

3. Is the evidence as it stands unambigous, or are there two or more
interpretations, and if so on what grounds do you choose one or the
other?


Obviously the further down any borrowability hierarchy the character is,
the more important the information under (2,3) becomes. But there are
situations where the character is both salient and suspect as a loan,
and the proper conditions under (2) occur(...) but where one can still
make a good case for the character not being borrowed. Assuming the
existence of such cases, it is then necessary to do a certain amount of
balancing of probabilities, as well as making purely methodological
decisions.

p.212

{regarind L uates , OIr faith, OE wopbora...}
What does one do with things like this? There are three obvious ways
they could come about;

1. The word represents an Indo-European root that was lost everywhere
except in one subfamily or small areal grouping.

2. The word was invented ex nihilo in the groups it appears in, either
in a `proto-' stage (dubious for Germanic/Baltic or Northwest IE) or
later; in the second case (as in `gold') it later diffused across a
Sprachbund.

3. The word was borrowed from some (vanished) local sub- or adstratum
family.
<><><><><><><><><><><><><><>----Language----<><><><><><><><><><><><><><><>
Copyrights and "Fair Use":     http://www.templetions.com/brad//copyright.html
"This means that if you are doing things like comment on a copyrighted work, making fun of it,
teaching about it or researching it, you can make some limited use of the work without permission.
For example you can quote excerpts to show how poor the writing quality is. You can teach a
course about T.S. Eliot and quote lines from his poems to the class to do so. Some people think
fair use is a wholesale licence to copy if you don't charge or if you are in  education, and it isn't.
If you want to republish other stuff without permission and think you have  a fair use defence, you
should read the more detailed discussions of the subject you will find through the links above."