Re: New Methodology on Analysis of Language Change

Joseph W. Murphy wrote:
>
> On Thu, 12 Jan 2006 23:03:59 GMT, Peter T. Daniels wrote:
>
> > Thanks again [for the data excerpts]. (You're spending an awful lot of
> > time on something worth nothing.)
> >
> > Yep, the methodology is identical to Ringe et al.'s (yet I didn't notice
> > anything by that team in their references).
> >
> > And their features are AFAICT entirely typological. You could do the
> > chart for 5000 languages and come up with a classification that had
> > nothing whatsoever to do with reality.
> >
> > In fact, they don't even claim to have tested it by using it either for
> > well-understood families, say IE, or for a random selection of
> > languages!
> >
> > Once again: Why does *Science* not use peer review for linguistics
> > articles?
>
> I posted it to get feedback and clarification from you and others here on
> the methodology employed. Ringe seems to have done the same thing with
> Indo-European -- only he used lexicology and not typology. I guess I'm
> doing it also to try to work through the way they did it in my own mind.
> Except that I'm still baffled because I don't understand anything about
> cladistics.

At least in the Ringe article published by Mair (1998), they used 20-odd
morphological and structural (i.e. non-lexical) features.

> Apparently, Dunn et al. did "test" their methods against that Austronesian
> languages in the same area Ross Clark mentioned (whose relatedness and
> classifications are supposedly well-known) and the results supposedly
> correlated.

I didn't notice that mentioned in what you posted.

> In the "Supporting Materials" posted as a companion to the article, things
> seem to have been done in the following way:
>
> 1. A questionnaire was compiled using the 125 typological factoids and
> each language was checked for the presence or non-presence of each factoid
> by linguists "with fieldwork expertise in the languages" or "from published
> and unpublished sources".
>
> 2. "Trees were then generated for each set of data using maximum parsimony
> methods, using the heuristic algorithm with taxa randomly reordered 100
> times over 10,000 bootstrapped datasets. The concensus tree from this

You retyped this, didn't you? Or did *Science* really spell <concensus>
and (5 lines lower) <realtively>? You're allowed typos, but *Science* is
supposed to proofread.

This is essentially the methodology -- at least, it uses the same
phrases -- that Jacques Guy demolished when they published the earlier
paper by Gray. So it's not just GIGO wrt data, but wrt method as well!

> bootstrap data provides us with an approximation of the relationships found
> in the dataset, and a measure of the robustness of particular hypothesized
> relationships. The level of homoplasy (convergent features independently
> innovated in different branches of the family tree) in the data is
> realtively high compared to biological genetic systems, caused by the
> relatively high (and probably variable) rate of change in languages, and
> because there is a level of functional interdependence in grammatical
> systems. Furthermore, the structural features used in the analysis are
> necessarily defined coarsely: more narrowly definded categories would
> provide higher apparent levels of support for clusterings of taxa, but this
> would be determined by the more-or-less arbitrary selection of the
> categories coded. Because of the relatively high level of homoplasy found
> in linguistic structural data, consensus trees are preferred to optimal
> trees as the basis for our phylogenetic hypotheses. To reduce the effect
> of homoplasy we used an a posteriori weighting of traits showing
> phylogenetic reliability; weights were derived from the rescaled
> consistency index value of each trait determined on the initial run"
>
> The above paragraph baffles me completely. I have no idea what is being
> said. Somehow, trees were derived. That's all I understand.
>
> The fact that Julian Scaff (our famous postmodernist) might have been
> involved in computer work at Nijmegen makes me nervous. The above-quoted
> second paragraph sounds like computer-generated postmodernistic drivel to
> me. Gobbledegook. I am totally lost.
--
Peter T. Daniels grammatim@xxxxxxx
.

Relevant Pages

  • Re: New Methodology on Analysis of Language Change
    ... the methodology is identical to Ringe et al.'s (yet I didn't notice ... > chart for 5000 languages and come up with a classification that had ... "Trees were then generated for each set of data using maximum parsimony ... Because of the relatively high level of homoplasy found ...
    (sci.lang)
  • Re: New Methodology on Analysis of Language Change
    ... time on something worth nothing.) ... the methodology is identical to Ringe et al.'s (yet I didn't notice ... chart for 5000 languages and come up with a classification that had ...
    (sci.lang)
  • Re: new here, my lang project...
    ... > oo is better suited for the large scale, eg, more as an organizational ... By that time industry in the West had gotten ... >>employing any methodology consistently and not using a methodology. ... Meanwhile languages like COBOL that everyone loves to hate ...
    (comp.object)
  • Re: The map of typological features
    ... I was merely pointing out that PTD's *specific* criticisms were ... I also agree with the majority of the points re methodology that you made in ... there just weren't decent grammars of any languages at all in many SAm ... red languages have any blue features or vice versa. ...
    (sci.lang)
Quantcast