Classifiers

[Jeff Shrum]

I am looking at some language data with classifiers.  I do not know, in the model that FLEx is built on, how to handle classifiers.  Noun classes are considered inflectional features in FLEx.  But for classifiers is it recommended to use inflectional features, inflectional classes, or something else?  There is no inflection involved in classifiers so it seems to me they should be handled another way.  Would anyone who has experience modeling or making a dictionary for a language with classifiers mind sharing how you recorded classifier information for nouns in FLEx?

 

Jeff Shrum

SIL International

Language Technology Consultant

7500 W. Camp Wisdom Road

Dallas, TX 75236

Office: 972-708-7400. Ext: 2030

Cell: 903-371-0297

 

[maxwell]

On 2017-01-25 11:22, Jeff Shrum wrote:
> I am looking at some language data with classifiers. I do not know, in
> the
> model that FLEx is built on, how to handle classifiers. Noun classes
> are
> considered inflectional features in FLEx. But for classifiers is it
> recommended to use inflectional features, inflectional classes, or
> something else? There is no inflection involved in classifiers so it
> seems
> to me they should be handled another way. Would anyone who has
> experience
> modeling or making a dictionary for a language with classifiers mind
> sharing how you recorded classifier information for nouns in FLEx?

Others may want to weigh in on this, but as one of the authors of the
original model (and realizing that things may have changed since then,
there may be other theoretical perspectives, etc.), I'll give my two
cents.

The term "(noun) classifiers" is used variously, as is the term "noun
classes". The latter can mean declension class (= inflection class),
which is presumably *not* what you're talking about. But the term "noun
classes" is also used for Bantu noun classes, which are quite different
from declension class. By definition, a declension class has no effect
outside the noun itself. That is, a declension class does not impose
any agreement on any other words--modifying adjectives don't agree with
the declension class of the noun they modify, verbs don't show
agreement, etc. As the term "noun class" is used in Bantu, however,
there is agreement outside of the noun itself (e.g. subject and object
agreement on the verb). In this respect, a (Bantu) noun class is much
more like a gender class in typical IndoEuropean languages.

As for classifiers, I'm assuming you mean classifiers that attach to
nouns (as opposed to numeral classifiers, as in Mayan languages). Noun
classifiers can be like noun classes in Bantu (for example) in the sense
that they may (but don't necessarily) cause agreement between themselves
(or the noun they attach to) and other words. But unlike typical noun
classes, in some languages a given noun can take more than one
classifier, with different meanings. For instance in Cubeo (a Tucanoan
language of Colombia), the noun stem xokɨ "wood" can appear alone, or
it can take a classifier -kɨ "tree-like object" to mean "tree", or the
classifier -we "flat object" to mean "board." (And in case you're
wondering, yes, this is like compound nouns, and in fact some stem+CLS
are clearly compound nouns.) If this is the kind of classifier you
mean, then I think you are correct that they're not like typical
inflectional affixes; they're more like derivational affixes.

Which of course brings up another question: is there a real distinction
between inflection and derivation? The FLEx model makes such a
distinction, but if you want, you can think of that distinction as more
an ease-of-use thing than a theoretical distinction. Engineering,
rather than science. Inflectional affixes in FLEx are arranged in
templates with slots. That makes it easy to get their positioning,
order and (many) co-occurrence restrictions right. If your classifiers
fit that model, then you can treat them in FLEx as if they were
inflection, without IMO having made a theoretical commitment. If on the
other hand your classifiers "stack" (you can have > 1 on a given word),
or if they grade into compound nouns (like Cubeo), then it would
probably be easier to treat them as derivational in FLEx.

So in sum--and with the caveat that I'm not sure exactly what you mean
by "classifiers"--I'm pretty sure you do not want to treat them as
inflection (declension) classes. You may or may not want to treat them
as inflectional affixes (bearing inflectional features). The
alternative would be to treat them as derivational (which they seem to
be in Cubeo).

Mike Maxwell
University of Maryland

[Beth-docs Bryson]

In languages where classifiers are separate from the words they classify, they could be modeled with a Lexical Relation in FLEx.  Maybe this relationship does not fit the standard definition of what a lexical relation is in linguistics, but it is one way to create links between words that are related in one way or another.

We ship FLEx with a Lexical Relation called "Classified Noun".  The reverse label is "Classifier for".  

Using Chinese as an example, the idea is that on a noun like lù 'road', there would be a notation:  clf.: tiáo 'CLF.long' indicating what its classifier is.

Then on the classifier itself, there could be a notation like:

clf. for: lù 'road', yú 'fish'

However, you may not want the classifier itself to list *all* the words that it can classify.  For some classifiers, this list would be huge.  The solution to this is to use the new "Unidirectional Lexical Relation", where you can link the words, but the relation only shows on one of them.  (This is only available in FW 8.3.x, and there are more details about how to use it in the Help system.)

Looking at this now, I realize there is a small bug in the use of the labels for this relation in FW 8.3.x.  Between 8.2 and 8.3 we have changed how "abbreviation" and "reverse abbreviation" are used.  This change happened okay for most of the lexical relations, but the classifier one needs a little more adjusting.  You can certainly edit the labels yourself in your own project.

This may or may not address what Jeff was asking about, but it is something worth being aware of.

-Beth

--
You are subscribed to the publicly accessible group "FLEx list".
Only members can post but anyone can view messages on the website.
To change your status, please write to flex_d...@sil.org.
--- You received this message because you are subscribed to the Google Groups "FLEx list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+unsubscribe@googlegroups.com.
To post to this group, send email to flex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/623231d58c0e0fb75201c0ea85c84ef5%40umiacs.umd.edu.

For more options, visit https://groups.google.com/d/optout.

[Jeff Shrum]

Beth,

 

Thank you, this is the kind of information that I was looking for.  Using a kind of lexical relation for classifier is one solution. How would this look in a printed dictionary?

 

Jeff Shrum

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+...@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/623231d58c0e0fb75201c0ea85c84ef5%40umiacs.umd.edu.

For more options, visit https://groups.google.com/d/optout.

 

--

You are subscribed to the publicly accessible group "FLEx list".
Only members can post but anyone can view messages on the website.
To change your status, please write to flex_d...@sil.org.
---
You received this message because you are subscribed to the Google Groups "FLEx list" group.

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+...@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/CAEj%3DFHq%2Bz6%3D-y1gj%2BtzyH67iGVVRC%3Dbs%2BCxHz3PGXzHZ723zAQ%40mail.gmail.com.

[Beth-docs Bryson]

I don't know if attached graphics will make it to the list, but I will try.

Here is what it should look like for (a) a classifier and (b) a noun that takes a classifier.

-Beth

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+unsubscribe@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/623231d58c0e0fb75201c0ea85c84ef5%40umiacs.umd.edu.

For more options, visit https://groups.google.com/d/optout.

--
You are subscribed to the publicly accessible group "FLEx list".
Only members can post but anyone can view messages on the website.
To change your status, please write to flex_d...@sil.org.
---
You received this message because you are subscribed to the Google Groups "FLEx list" group.

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+unsubscribe@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/CAEj%3DFHq%2Bz6%3D-y1gj%2BtzyH67iGVVRC%3Dbs%2BCxHz3PGXzHZ723zAQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--

You are subscribed to the publicly accessible group "FLEx list".
Only members can post but anyone can view messages on the website.
To change your status, please write to flex_d...@sil.org.
---
You received this message because you are subscribed to the Google Groups "FLEx list" group.

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+unsubscribe@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/6770d6df6391a09722e9dfa2a17bc0f4%40mail.gmail.com.

[Paul Unger]

This looks helpful to me, too. I think... In Doku (Oceanic), we have two possession classifiers: /ni/ 'general CLF' and /gha/ 'food, kin, and other close-to-the-heart items CLF'. These are inflected for possessor and number (dual, paucal), for example:

/na ni-gua na thinaghe/ 'ART gen.CLF-PS:1SG ART canoe; my canoe'
/na gha-mua na vudi/ 'ART food.CLF-PS:2SG banana; your banana(s)'
/na ko-gha-dira na vudi/ 'ART DL-food.CLF-PS:3PL ART banana; their two's banana(s)'
/na tu-no-da na kei/ 'ART PAU-gen.CLF-PS:1PL.IN ART basket; our(inclusive) basket(s)'

To complicate things(!), you can see the variant in the fourth example: /ni/ is used for 1sg, 2sg, 1pl.ex, and 2pl, and /no/ for 1pl.in and 3pl. 3sg is /ne/, which makes a bit more sense when you consider the morphophonemic process that affects 3sg /gha/: when the 3sg possessive suffix /-a/ joins to a stem ending in /a/ the two /aa/s raise to /e/ (there's vowel harmony, too, but since /gha/ is one syllable, we don't need to bother with that here). So /gha/ + /-a/ -> /ghe/.

Is this something FLEX could handle? I've never been able to get the parser to work for my data--there seem to be enough of these kinds of things that I guess I've given up... [e.g., the possessive suffixes also join to inalienably possessed nouns; and the output from the 3sg possessive morphophonemic process (i.e., /-e/) joins to (some) cardinal numerals to derive ordinals; etc.] Too many 'variant' and 'some, but not all' situations. But if anyone has suggestions for something like this, I'd be open to hearing them!

Thanks.

To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+...@googlegroups.com.

To post to this group, send email to flex...@googlegroups.com.

[Jeff Shrum]

Paul,

 

When grammar templates are not adequate, (i.e. they allow two classifiers to match with the same possessive marker) then you need to make ad hoc rules that restrains the parser from allowing the unwanted match to your template.  You can find information on ad hoc rules in Help.  In general ad hoc rules should only be needed sparingly, because a large number of ad hoc rules is a sign of analysis that does not adequately describe the grammar.  (Having said that, how many are too many, well who can say exactly.  Just take it as rule of thumb that may be helpful to you.)  If a difference is morphologically conditioned then define natural classes for you language and make environments to tell the parser when to use the allomorphs.  Unfortunately, the parser cannot use any rules that cross word boundaries so making a phonological environment rule will not help in cases where more than word is involved.

 

Jeff Shrum

SIL International

Language Technology Consultant

Dallas, TX

USA

 

From: flex...@googlegroups.com [mailto:flex...@googlegroups.com] On Behalf Of Paul Unger
Sent: Wednesday, April 12, 2017 6:18 PM
To: FLEx list <flex...@googlegroups.com>
Subject: Re: [FLEx] Classifiers

 

This looks helpful to me, too. I think... In Doku (Oceanic), we have two possession classifiers: /ni/ 'general CLF' and /gha/ 'food, kin, and other close-to-the-heart items CLF'. These are inflected for possessor and number (dual, paucal), for example:

You can join this group by going to http://groups.google.com/group/flex-list.

---
You received this message because you are subscribed to the Google Groups "FLEx list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to flex-list+...@googlegroups.com.
To post to this group, send email to flex...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/flex-list/73d22ba0-57c2-4f08-9590-1a22ec3b5794%40googlegroups.com.

[Paul Unger]

Hi Jeff,

Thanks for your reply. I'm afraid you're a number of steps ahead of me, though...

Are you saying that the ad hoc rule would apply to the classifiers? Or to the allomorphs of the 'general CLF': /ni/, /ne/, /no/? Or to the possessive markers (which also join to inalienably possessed nouns, a noun-like preposition, the reflexive marker, nominalised verbs, and a small set of 'stative' verbs [s.t. is like another; s.t. is fit, meet, appropriate, enough to another])?

I've taken a few runs at the "Introduction to Parsing..." document (in Help > Resources), and I've made a few templates, but I can't seem to constrain it enough to satisfy the parser... Sometimes it seems that I don't know which end of the stick to hold, so to speak. In fact, I just tried to run the Parser ("Try a word") and I've crashed FLEx twice. So whatever I've got in there isn't happy... :-(

Paul