LINGUIST List 5.524

Fri 06 May 1994

Sum: Computational models of parameter setting, Black holes

Editor for this issue: <>

Directory

1. Alison Henry, Computational models of parameter setting
2. , black holes: summary

Some time ago I posted a request for information on recent work on computational
 models of parameter setting. Thank you to all who replied: Bill Turkel, Carole
 Tenny Boster, Paul Chapin, Bill Idsardi, Jeff Lidz, David Leblanc, Simon Kirby,
 Shalom Lappin, Andi Wu, Beatrice Santorini, Tod Wareham, Philip Resnik, Eleanor
 Olds Batchelder, Kevin Broihier (apologies if I have omitted to mention
 anyone). Here is a list of the main references suggested.

Clark, R & Roberts, I (1993) A computational model of language learnability and
 language change LI 24 (299-345)

Clark, R (1992) The selection of syntactic knowledge Language Acquisition 2
 (83-149)

Wexler, K & Gibson 'Triggers" (to appear in LI)

Wu, A (1994) UCLA PhD dissertation

Dresher, Elan & Kaye (1990) A Computational Learning Model for metrical
 phonology Cognition 34 (137-195)

Turkel, B a paper on genetic algorithms as a model of acquisition in
 Optimality Theory, available by anonymous ftp from hivnet.ubc.ca

Alison Henry

Thanks to everybody who replied to what was not as inocuous a
question as I thought. I have probably thanked everybody
individually, so if your name doesn't appear in the summary it's not
through ingratitude. The 'proper' summary should wait till I've
digested all the responses and done all the assigned reading, but I
may not live that long.

1. My original wuery was whether there were other pheromena like the
English his-genitive (John his book, etc.), where an affix apparently
climbed the grammaticalization hierarchy the wrong way and became a
word. That is, is grammaticalization (or the mere fact of being
'grammatical') a black hole or a sink in the dynamical systems theory
sense. It certainly is the latter in the sense of 'flow lines'
converging on it.

Lots of people questioned my analysis: the mere fact that the his-
genitive is later than than the -'s genitive doesn't mean it's a
reanalysis of it (the post hoc ergo propter hoc fallacy). This of
course is true.

Also many respondents (especially Hartmut Haberland, Tom Shannon,
Martin Haspelmath) pointed out a postposed pronominal possessive in
other Gmc languages, e.g. the German meinem Vater sein Haus with
dative of the first possessive, Dutch mij vader zijn huis, and the
further reduced Afrikaans clitic type my vader se huis. None of these
could be from an old affixal form, but are new alternative creations,
what Meillet would call 'renouvellement' rather than
degrammaticalization. The fact that -'s has the phonological shape it
does then may trigger the choice of 'his' as the postposed form; this
is also a good story because at this stage his is not only masculine
but neuter (its doesn't arise till much later).

2. As pointed out by Martin Haspelmath and Andrew Carstairs-McCarthy,
my analysis of the modern -'s genitive as an affix is faulty: it's
now best taken as a clitic. If this is so, then in fact it is itself
a better example, since the modern -'s is itself then a clitic that
has exited from the affix sink (it's been a suffix as long as anybody
knows). In fact the affix-to-clitic move is so well known and patent
that I wonder why I didn't think of it. Until well into Middle
English times what Jespersen calls the 'group genitive', i.e. '[the
king of England]'s' nose did not exist, bu the usual type was '[the
king]'s nose of England'. In Old English the usual structure, before
the use of the of-possessive would have been 'the king's nose
England's'; Jespersen (Modern English Grammar 17.1) cites one that
would parse as 'Aethelwulf-gen.sg daughter West-Saxon-gen.pl king-
gen.sg.', i.e. in the later form [Aethelwulf King of the West
Saxons]'s daughter'.

It is therefore clear that the modern status of the genitive is the
kind of case I was looking for.

3. Several people (Linda Colemean, Kevin O'Donelly) have pointed out
the case of the Modern Irish -muid, which apparently is a verbal
concord on its way to becoming a fre pronoun, and a number of Uralic
cases appear in Lyle Campbell's paper in the Traugott & Heine 1991
collection. Hartmut Haberland has also suggsted Greek ksana 'again',
which derives from a complex verbal prefix.

4. Somewhat similar phenomena (but in a sense less 'erious') have
been suggested by others: e.g. what might be called neoclassical
affixoids > words in collocations like 'I'm anti that', 'Your
argument is becoming very meta' (Stephen Spackman), or '-ism' as a
noun (Robert Beard); and of course there's always 'pros and cons'.
Others suggested bus < omnibus as a kind of cross-linguistic example,
but this is more in the nature of learned word-play to begin with or
deliberate clippling (like mob < mobile vulgus). But this kind of
thing (and metanalysis in general) do show an ability and willingness
to segment the unsegmentable, and reify the unreifiable. ALso of
course pseudo-affixes like -kini (bikini = bi-kini, hence monokini
for a topless one), similarly '-tique' = 'shop'.

5. I was not trying to claim of course that the 'proper' hierarchy is
not the commonest direction, and for good reasons. Grammaticalization
is a normal attractor, annd like all point attractors it takes a
special kick to get something out once it's fallen in. But is not a
black hole in the strict sense. What I was interested in is how
common the phenomenon of reversal is, and it seems that the answer is
as I suspected not very. But it can't be discounted.

6. As a final pint, along the lines of something brought up by Jason
Johnston, how does one treat movement along a sequence where say a
thematic noun-class marker that was once (possible) a derivational
affix becodmes more 'informative' by turning into an inflection?
E.g. the -Vn- formative that occurs in Latin n-stems like homo, hom-
in-is is in Latin (whatever it might have been earlier) mainly a
declension marker; but in Germanic this thing gets reconstrued as a
case/number morph. So in the cognate OE guma 'man', gen/dat/acc sg,
nom/acc pl guma-n, what's left of the declension marker has syntactic
functions. Is an inflection more 'lexical' than a class-marker? I'd
say yes. Of course again the opposite is comoner, e.g. if as is
commonly believed the -s- in cleanse (< OE claen-s-ian) is some kind
of descendant of the IE -s- aorist marker.

I have found the whole thing very interesting and informative, and
would be delighted by further public or private discussion.

Thanks again to everybody who replied, and whose replies occupied a
rather small amount of disk space in relation to the queries about
the 'real story' of the South African election, which occupied more
of my time lately than linguistics. What you got on CNN and Sky News
is as close as you'll get for the moment.

Roger Lass