No Subject

[Logical Language Group <lojbab@ACCESS.DIGEX.NET>]

To: Leo Connolly
==========================================
Responding very late to this, Leo, and hence not to the newsgroup,
I am sending separately you a copy of my response to Ken Shan who you are
responding to in this message.  I'm repeating your full message since you
may not have it handy.

>From: connolly@msuvx1.memst.edu
>Subject: Re: VSO languages? (was: Prefix languages?)
>Date: 16 Jun 93 10:25:13 -0600
>Organization: Memphis State University
>
>u7911005@cc.nctu.edu.tw () writes:
>> connolly@msuvx1.memst.edu said:
>> : The problem is considerably more complex.  Part of the problem is that
>> : linguists -- including me -- are too often tempted to make the traditional
>> : [...]
>> : What of subjects and objects?  Well, though subjects in languages of the
>> : normal Western European type, such as English, do tend to make the doer
>> : of the action (if there is one) the subject -- but they don't have to;
>> : it's perfectly OK to say "Charlie was arrested by the sheriff."
>>
>> Here the subject and object of the sentence are reversed by the
>> use of passive voice.  Clearly, the meaning of "Charlie was arrested
>> by the sheriff" and "Charlie arrested the sheriff" is different.
>> I don't know if linguistics consider "Charlie" or "sheriff" the
>> "subject" of the passive-voice sentence, but I think the "sheriff"
>> is the "true" subject.
>
>I agree.  Many languages, including English, permit alternate subjects
>with special verb forms.  These are indeed true subjects.
>
>> : What's a subject?  Very simple: a noun or pronoun which
>> : in some particular language gets this special treatment reserved for one
>> : noun or pronoun per clause -- whatever that treatment may be.
>>
>> I see.  When we analyze a sentence using predicate logic (i.e.,
>> arrest(sheriff(), Charlie())), I think we can take everything as
>> a function, so:  A noun would be a function without arguments;
>> and a verb or an adjective would be a function with one or more
>> arguments.  And the "subject" would be the "first argument" of
>> the verb function, which (as you pointed out) doesn't have to
>> be the "active argument" at all.
>
>Well, it would be the special treatment accorded a first argument that
>makes it the subject.  Things do get confusing when, in instances like
>the Spanish sentence below, what by rights "should" be the first
>argument (and actually is in English and many other languages) is
>treated in part like an object, while still retaining some subject
>properties.
>
>But I do agree that at deep structure (or whatever you want to call it),
>there are no subjects, only arguments of a predicate.  Subjects are
>created by syntactic processes, which is why alternate subjects are
>possible.  This, by the way, is the standard treatment in Case Grammar,
>of which I am an adherent.
>
>> (Here, some particular "features" of English are making thinking
>> generally about all kinds of languages a bit more difficult.)
>
>Absolutely.  One reason is that English does not permit the sort of
>split treatment seen in Spanish where one NP has some alleged subject
>properties, while another NP has the rest.
>
>> : Finally, someone else claimed in a post that the notion of "subject" may
>> : not apply in "ergative" languages such as Basque.  Ergative languages have
>> : subjects all right, but they identify them on different principles.  When
>> : a verb has a "doer" and a "receiver of the action", we would make the
>> : doer the subject -- they make the receiver the subject and give it exactly
>> : the same treatment they would give to the only noun in sentences such as
>> : _The children are running_, _The woman is tired_ etc.
>>
>> Would that be a language with the "passive argument" always
>> being the "first argument" to the verb?
>
>This notion has been around for a long time.  See Uhlenbeck, _Het
>passieve Werkwoord..._, for a classic exposition (in Dutch,
>unfortunately).  I don't think it's right.  In passive sentences, the
>normal subject is somehow demoted, losing all subject properties and
>syntactic "salience".  The Relational Grammarians say that it has become
>a "cho^meur" i.e.  'unemployed person'.  In almost all ergative
>languages this is not true:  the agent that we would make the subject is
>still syntactically salient, participating in numerous syntactic
>processes, often controlling reflexivization and sometimes even verb
>agreement (though not usually).  So despite the similarity that you
>mentioned, the syntax of ergative sentences and passive sentences is
>quite different.
>
>> : So: is a distinction between noun and verb necessary?  Apparently yes; at
>> : least, all languages make distinctions, though some much more clearly than
>> : others.
>>
>> As I said above, the "everything's a function" (i.e., lambda
>> calculus) concept, when applied to designing (artificial) languages,
>> can result in no distinctions between nouns and verbs being needed.
>> For example, in a postfix artificial language based on lambda
>> calculus (where "postfix" means "all the function arguments (if any)
>> comes *before* the function name), the sentence "The sheriff arrested
>> Charlie" would become "Sheriff Charlie arrest."
>> (Of course, the tenses of verbs are being ignored here.)
>
>It don't think that's adequate.  It is necessary to distinguish the
>arguments of the verb-function not as first, second, third etc., but as
>agent, experiencer, patient, etc.  The reason is simply that not all
>first, second, third arguments bahave alike.  In many languages, verbs
>with experiencer as first argument differ syntactcally from those with
>agent as first argument.  In English, the experiencer-verbs notoriously
>do not have natural imperative or progressive forms; in German, they do
>not passivize.  Compare:
>
>A[1] = Agent:   I am listening to my mother.
>                Listen to your mother.
>
>A[1] = Exper.:  *I am hearing my mother.
>                *Hear your mother.
>
>Given the extreme similarities in meaning, these constraints cannot be
>formulated in semantic terms.  But meaning is related:  verbs with
>obviously active arguments that do not perceive (or benefit) are not
>subject to these restrictions, but many verbs of perception (and
>benefit) are.  That is, some of these must be indexed as having A[1] =
>Agent, others as A[2] = Experiencer.  (This too is a Case Grammar view,
>though not the "classic" one.)

In Lojban there is no syntactic distinction between arguments based on
case, although there is an optional case-tag system used most commonly
to add arguments that are not normally considered essential to the
predicate itself (tense and location, observers, motivations and causes,
etc.)  These can be used on the arguments defined to be part of the
predicate but this would be abnormally marked, since the tags aren't
necessary.  I have not noticed that any predicates suffer restriction in
Lojban usage thus far, though as I noted, some speakers are starting to
turn modal predicates like "possible" into VSO order even at the expense
of the more highly marked form.  A couple of people who do this and are
linguistically aware explain this in terms of moving a heavy clause to
the end of the sentence to reveal the predicate more quickly and easily.

>> Some kinds of ambiguity in English can be avoided this way.
>> For example, the sentence "They are flying airplanes"
>> have several different meanings in (simplified) predicate logic:
>> "flying(they, airplanes)", "be(they, flying(airplanes))", etc.
>> In the abovementioned postfix language, these meanings would
>> become: "Fly airplanes they" and "They airplanes flying be".
>
>The same is true of any system which distinguishes deep and surface
>structure.  Ambiguity, Chomsky (1957) said, results when different deep
>structures can produce identical surface structures.  Even his examples
>were similar.

It has been argued that speaking a language like Lojban is very much
speaking a language whose deep structure matches the surface structure.
What this means for linguistic theories if Lojban develops 'native
speakers' i.e. children who can learn the language through natural
processes, is to me an interesting question.  (Any such children will of
course be learning some other language at the same time, and there may
be some intersting effects of cross-language interplay that affect
acquisition.  My own two kids, adopted from Russia 10 months ago, are
ages 5 and 7 and still learning English, but are starting to acquire
individual Lojban words as well - though no sense of multi-word syntax
yet.)

>> But (just as in any other language) the only way to *completely*
>> avoid ambiguity is to use parenthesis-like devices to denote the
>> "number of arguments" to each function.
>
>Of course, these are at deep structure.  I've never heard a parenthesis,
>so the problem of surface ambiguity (the only kind known to exist)
>remains.

In Lojban the parentheses that group the structures are explicitly spoken
except when not necessary because no ambiguity results under the formal
grammar (defined under LALR1 rules with YACC).

The acquisition (thus far only in adults of course) of the rules for
when elision of parentheses is permitted and/or desirable may prove to
be revealing of some aspects of how our brains process the predicate
structures that comprise the language.  People sometimes include
disambiguation terminators when not necessary, and I think there are
significant patterns in the usage, though I haven't yet determined what
they are.

>> The problem is that,
>> given a certain function, there might be different number of
>> arguments to it, and there are even "optional arguments" identified
>> by "keywords/prepositions".  For instance: "He gave." "He gave it."
>> "He gave it to her."
>
>Well, that's a problem in any system.  Some formalisms may handle it
>better than others, but you'd be no worse off than the rest of us, I
>think.

As I said, we have for each defined predicate some fixed number of
mandatory arguments, and a mechanism through case tags of adding
additional optional arguments.  There are also mechanisms that define
'new' case tags via relative clauses, though these are rarely used.
Lojban has a lot of bells and whistles built in specifically to find out
which things are necessary to speakers, which are optional, and which
just cannot be acquired naturally.

>
>[stuff omitted]
>
>> Here I see three kinds of distinctions, all getting confused with
>> English:
>>     - subject/object
>>     - first argument/other arguments
>>     - before verb/after verb
>
>Yup.  But they get confused everywhere else too if we're not careful
>with our definitions.  In particular, the word order problem is relevant
>in many languages even for two or more arguments on the same side of the
>verb.  Indeed, languages which are verb-initial or verb-final are even
>more confusing in this regard than English.  To mess things up still
>further:  many languages permit the theme:rheme (aka "topic:comment")
>distinction to affect such matters as word order and (to some extent)
>choice of subject.  You mnight also want to read up on "unaccusativity".
>That's a notion developed in Relational Grammar and accepted by lots of
>GB people (I'm neither!).  The problem concerns subjects -- real
>subjects -- which act like direct objects (details vary greatly from
>language to language).  While I don't buy the RG or GB explanation, that
>they actually *are* direct objects at deep structure, the phenomena are
>very real and must be accounted for.  That too seems to require indexing
>of the sort I proposed above.  And if you don't like my explanation, or
>theirs, the problem must still be confronted and resolved somehow.
>
>--Leo Connolly

As I say in my other post, I can't see any distinction between direct
and indirect object in Lojban, though perhaps one could define a
case-tagged argument as an "indirect" one.  However, that would still
leave several direct objects in some predicates (Lojban klama = go() has
5 defined arguments, hence a subject and 4 direct objects, by this
analysis.)

Lojban also has a topic comment structure, though only a few people know
the language well enough to use it comfortably.  It corresponds to the
logical prenex struture of symbolic logic.  If a definite argument is
put in the prenex, it acts like a topic, making the entire sentence a
comment.  Grammatically, the topic comes before all other sentence
parts, including a differing subject, because of the prenex model.  I
don't believe it would be hard to add a non-initial topic to the
language, though it would have to be pre- and post- marked except at the
end of the sentence where the ending could optionally be elided.  (the
current topic structure is marked by a word that separates the prenex
from the rest of the sentence).

lojbab