LINGUIST List 13.2404

Sat Sep 21 2002

Disc: RE: Do We Need a Replacement for *(...)

Editor for this issue: Renee Galvis <reneelinguistlist.org>

Directory

1. Peter T. Daniels, Re: 13.2381, Disc: Do We Need a Replacement for *(...)
2. Tim Stowell, Re 13.2369, Disc:Do We Need a Replacement for *(.)
3. Chris Johnson, Re: Do We Need a Replacement for *(...)?

Message 1: Re: 13.2381, Disc: Do We Need a Replacement for *(...)

Re Linguist 13.2369

Jim McCawley (1993 LACUS Chicago) suggested a linguistics of
grammaticality judgments -- the data wouldn't be the judgments
themselves, but the variation in whether particular sentences were
judged grammatical by all, some, or no consultants. (He reported on
disagreement among the students in classes about particular examples.)

BTW when I tried to explain this at the question period at an NYU talk
the notion of variability in grammaticality judgments was ridiculed by
more than one person.

- 
Peter T. Daniels 

Joost Kremers points that, while (*X) is intuitive, *(X) is not. His
proposal to introduce (+X) is not unreasonable, but I foresee
practical problems in its implementation, and I think that his
intended meaning of ''+'' is different from what he
suggests. Christopher Bader's proposal to solve the problem by using
only comparative judgements (indicated by >) strikes me as a red
herring. I therefore suggest that we leave things as they are.

First, as Kremers acknowledges, the existing literature uses *(X), and
if we want to encourage new linguists to read the old literature, we
will still have to teach them *(X). Moreover, it is unlikely that (+X)
would be universally adopted, and if both conventions were in common
use, we would have to teach people the meaning of both (+X) and *(X).

Kremers remarks that ''what our notational toolbox generally lacks is
a ... character that indicates that a given phrase is
grammatical....We can use the plus character ''+'' for this marker.''
Actually, there are two existing notations for grammaticality (or,
more accurately, for acceptability). The first is the absence of any
special marking: the default value for an example is [+acceptable]. A
check mark is also sometimes used as an overt marker, but it is not a
standard ASCII character, so Kremers's proposal to replace it with +
has potential merit.

Still, his proposed usage of + in (+X) does not really convey
[+grammatical] or [+acceptable], but rather something like
�Eurooeobligatory�Euro. If + were to indicate
[+grammatical], then the string �EurooeY (+X) Z�Euro would
only be naturally understood to convey that YXZ is grammatical iff it
includes X if the default value of an example were [-grammatical]:
omission of +X would entail omission of [+grammatical]. On the other
hand, if + were to mean ''obligatory'', then the inclusion of
parentheses in ''Y (+X) Z '' would be anomalous, at least if
parentheses indicate optionality, since it would mean that the example
may ''optionally include an obligatory element X'', which of course is
senseless. We could drop the parentheses, but then we would need to
include some kind of bracketing to mark the boundaries of the
obligatory sub-string.

Christopher Bader correctly draws attention to the implicitly
comparative status of grammaticality markings, but his proposal solves
Kremers's problem by brute force. The problem arises only because
syntacticians like to avoid citing two independent examples which
differ only in the presence or absence of a sub-string; we have
adopted the habit of combining such examples by means of
parentheses. The conventions of (*X) and *(X) --or (?X) and ?(X)--
were introduced to enable us to combine examples with different levels
of acceptability. Of course we can avoid Kremers's problem by
abandoning this habit, citing both examples in full, each marked with
the appropriate level of acceptability. This is essentially what
Bader's proposal does, though the issue is confused by his additional
advocacy of comparative judgements in place of absolute judgements.

I agree that judgements are often implicitly comparative, but there
are problems with abandoning absolute judgements entirely. For one
thing, one advantage to having an arsenal of notations to convey
varying degrees of absolute acceptability is that there is a real
difference between a marginally acceptable example and a completely
unacceptable one. Nuanced judgements can be conveyed conveniently with
combinations of markings such as **, ?*, ?, etc. There is no easy way
to convey these distinctions with > alone. One might think that this
can be achieved by a ranking of A > B > C, where B has the status of ?
and C has the status of *, but this obscures the distinction between a
3-way contrast of A vs. ?B vs. *C on the one hand and A vs. *B vs. **C
on the other. I can think of ways of trying to fix this while still
using only >, but none of them are very attractive.

At 13:47 +0000 9/19/02, Joost Kremers <j.kremerslet.kun.nl> wrote:

>Being a syntactician, I regularly encounter the markers "*(...)" and
>"(*...)" to indicate obligatory inclusion and exclusion, respectively.
>For example:
>
>1) I see *(the) car
>2) I see the (*a) car
>
>(1) indicates that the phrase is ungrammatical if "the" is left out,
>whereas (2) indicates that the phrase is ungrammatical if "a" is put
>in.
[SNIP]

I agree that the *(...) notation is sort of insane. You're 'negating'
the optionality of the element. Coming up with compositional meanings
for '*' and '(...)' isn't that obvious.

I don't really see the (+...) notation as being much better. It does
lessen the chance of confusion - *(...) and (*...) are easy to mix up,
leading to several minutes of head-scratching - but it still puts the
expression of optionality in your grammar.

To me, the obvious solution is "I see {the | a | *0} car". This way
you sort of 'nominalize' the optionality of the word; an empty
position is just one of several choices that may make the sentence
grammatical or not. This notation generalizes to other situations: "I
like {all | *every | 0} cars."

(It seems like there may be an interesting point here about
categorematic vs. syncategorematic meanings, the philosophical virtues
of nominalization, and maybe even metalinguistic negation, but my
mind's too frazzled to figure it out.)

I'm pretty sure I've seen this sort of notation before.


>So what do you think? Is it a good idea to replace *(...) or do you
>feel it is not necessary? And would it not be a bad idea to replace
>something that has already been in use for (I guess) several decades,
>possibly creating some problems for future generations of linguists
>when they want to read the older literature? Is it even feasible to
>replace something that has been used for so long?

I wouldn't worry about this too much. After all, linguists already
have to decode ancient formalisms (e.g., from the 70's) every once and
a while. You can generally figure this out from context: if the
surrounding text is trying to explain why 'the' is mandatory in (1),
you'll eventually figure out what *(...) means. It may be confusing
to linguists of the future, but it can be pretty confusing to
linguists today. Backward compatibility shouldn't stand in the way of
clarity.

As for the OT-style notation, that looks like it would work well for
short examples, but it would be problematic for multi-line examples,
possibly with interlinear glosses. The point of notation like (...),
*(...), etc. is to represent the grammaticality (which is, as Dan
Everett points out, a problematic concept) of several sentences at
once.

(Of course, you could use something like this to indicate gradations
in grammaticality economically: "I saw {the > 0 > the a} car.")


- 
Chris Johnson