[YG Conlang Archives] > [engelang group] > messages [Date Index] [Thread Index] >

[Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: [engelang] Re: [jboske] LoCCan3 development ideas.

What follows is a little long-winded and may have some typos, but it contains a sketch of the syntax and semantics of a two-place "serial-predicate" language derived heavily from Richard Morneau's MTIL monograph, possibly adoptable for LoCCan-like languages.

On Thu, Aug 16, 2012 at 8:47 AM, And Rosta <and.rosta@hidden.email> wrote:
 

Mike S., On 16/08/2012 01:33:


> Pentadic predicates? I would keep predicates lean and mean. There is
> really no need for bloat-happy hijinks /à la/ Lojban. The majority of

> predicates are going to have 1 or 2 inherent places, and very few
> need more than 3. I would argue that _all_ predicates can be
> decomposed into predicates of no more than _2_ places. Take "give"
> for example, with 3 inherent places. This can be represented by two
> predicates syntactically combined:
>
> xfr: Ax transfers Px
> kl: Pk passes to Tk
>
> so prna xfrai so mlti kliu ro vrbu.
> Some person (Ax) gave a cat (Px=Pk) to every child (Tk).

I agree that increase in the adicity causes exponential increase in complexity. In Livagian I had one ogdoadic (iirc) predicate -- it was something like "x1 is the event of x2 being in ordinal position x3 in set x4 defined by property x5 ordered by property x6 with cardinality x7", maybe it was hebdoadic not ogdoadic -- and it required a ghastly amount of baroque machinery to accommodate it.

Regarding "event x1", I assume that event predicates are formed regularly as Lojban NU expressions are formed?  Being a member of a set is not an event in the Lojban (or English) sense so I'm guessing that we're talking about a dynamic set-like entity, i.e. a set-function on time.  I probably misunderstand this, but here's my interpretation of how your hebdoad works:

I saw on television today the event (=x1) of John (=x2) becoming the third (=x3) (in set =elided x4 defined by being a) person to walk on Mars (=x5) (ordered obviously by time =elided x6) (with cardinality obviously now 3 =elided x7).

I find that x4 and x5 are really the same thing said different ways; the rest of the places are essential to the meaning, but as a whole it's too unwieldy: x3 and x6 are closely related to one another and might be managed by a separate predicate linked to x4/x5; likewise x7 is inherent in the x4/x5.  In software terms this thing needs to be refactored.


But not all decompositions seem to adequately capture the sense of the gestalt. E.g. "X says Y to Z". And decomposition is itself cumbersome, of course.

Why not "X tells Z Y"?  Either way, as you're aware, there will be times when you want to remove or demote the second argument.  English uses a bag of methods including unannounced valency changes, Morneau uses an anti-passive or anti-middle voice marker, and Lojban uses FA or {zo'e} or (rarely) multiple SE.  All of these things work (not so logically in English of course), but maybe the least cumbersome approach is to not have to skip what you don't add in the first place (see below). 

 
> If all predicates are kept to 2 places, then we will be able to
> afford suppletive forms for the inverse/passive and reflexive of
> common predicates as you suggested. The rest can be formed regularly
> using particles. The number of possible operations is relatively
> small if one insists on 2-place predicates; it quickly starts to get
> complicated if you go even to 3.

I agree that even with 3 places things get more complicated, but reducing to only dyadic leads to verbosity. E.g. "X says Y to Z" could be "X is talker in event W, and Y is message in event W, and Z is addressee in event W". Likewise the megapredicate I mentioned above could be restructured so that instead of one predicate with 7 arguments, you could have six dyadic predicates each of which have an argument corresponding to the x1 and the other argument corresponding to one of x2--7.

Am I to understand that every predicate in your system has to have an argument for event?  In that case reducing predicates to only two places would probably make it not only verbose but unworkable.   I am uncertain of the function of your event arguments and why they can't be grammaticized more efficiently, but I am curious to find out more about it, so please feel free to shed some light on that if you would.  If these events have all the related tense and modal logic worked out behind the scenes, I would especially like to see your notes. 

I see the triad "X says Y to Z" as most conveniently decomposed into dyads strictly among X, Y, and Z,  namely "X expresses Y" and "X addresses Z".  Here's the basic idea:

{say}: X says/expresses proposition/narrative Y
{tell}: X tells/addresses/communicates to audience Z

John {tell} Alice.
John tells Alice.

John {say} he is going to the store.
John says that he is going to the store.

John {tell} Alice {say} he is going to the store.
John tells Alice that he is going to the store.

John {say} the story {tell} Alice.
John tells the story to Alice.

Alice {inverse-tell} John.
Alice is told by John.

Alice {inverse-tell} John {say} the story.
Alice is told by John the story.

Alice {past} {inverse-tell} {say} the story.
Alice was told the story.

 
> By the way, {zi'o} will never be needed because bogus places simply
> won't be admitted into predicate semantics.

I doubt zi'o has ever had any fans.


> I think the best way to link the place structures of syntactically
> combined predicates is by simply adopting Richard Morneau's system of
> three archetypical thematic relations called agent (A), patient (P),
> and theme (T) (the last Morneau calls "focus"). These can represented
> by vowels a/i/u. Clauses can be composed of serial predicates as in
> the example above, or you can derive case tags like Morneau if you
> prefer; there is only one agent- and/or one patient-arguments per
> clause, each of which is governed by any number of co-predicates of
> the correct types; there are any number of theme-arguments, each
> governed by exactly one co-predicate. In my tinkering I have found
> that there are only four open-class predicate types needed: (A,P) "A
> does something to P", (A,T) "A does something using T", (P) "P is
> something", and (P,T) "P has something to do with T". There is also a
> need for a closed class of coordinating particles with arbitrary
> valency (P, T1, T2, T3,... ) but I'll leave that out for now.

Could you say a bit more about the syntax of serial predicates, and about why you need four (rather than just one) open classes?

The short answer is that the serial predicate syntax and corresponding semantics under consideration do not treat all predicate slots as interchangeable, which is what a one-open-class system like Lojban's requires.  Four appears to be the smallest number of open classes required to make this version of Morneau's system work. A description of the system follows for your curiosity.

The syntax of this version is basically SVO and SV.  As mentioned, an open class basic predicate has one of four types of argument structures, (P), (P, T), (A, P), (A, T).  The A refers to an agent slot, P to a patient slot, T to a theme slot.  Argument structure type is lexically determined.

The agent and the patient each appears at most once per clause, filling all serial predicate slots of corresponding type.  Each theme on the other hand is an argument only of the predicate that introduces it.

Some vocabulary for examples:

{roll}: P rolls
{damaged}: P is damaged
{enter}: P enters/goes into T

{make}: A does something to P   // the generic (A, P) predicate
{push}: A pushes (on)/presses against P

{make-roll}: A rolls P                   // regular derivation
{make-damaged}: A damages P   // regular derivation
*{make-enter} A puts P into T       // illicit derivation due to arity constraint under consideration
*{make-push} A makes A push P  // illicit derivation due one-agent-per-clause constraint

{use}: A uses/does something using T    // the generic (A, T) predicate
{use-forklift} A uses a forklift on P           // regular derivation


Here are two predicates with patient slots, in this case both in the subject position, one intransitive and one transitive.

The barrel {roll}.
The barrel is rolling.

The barrel {enter} the room.
The barrel is entering the room.


When two predicates with patient slots occur serially, the patient argument appears only with one, but nevertheless is understood to fill the patient slots of both, the resulting combination is taken as a single coherent event or state.

The barrel {roll} {enter} the room
= The barrel {roll} + [the barrel] {enter} the room
The barrel is rolling into the room.


Predicates can be reordered freely without changing semantics:

The barrel {enter} the room {roll}.
= The barrel {enter} the room + [the barrel] {roll}.
The barrel is going into the room rolling.


When any predicate of type (A, P), such as the generic {make}, is introduced into the clause, every patient-oriented predicate (P) and (P, T) acquires an agent slot *semantically*, but not syntactically.

John {make} the barrel.
John is doing something to the barrel.

John {make} the barrel {roll}
= John {make} the barrel + [A:John] [the barrel] {roll}
John is making the barrel roll.

John {make} the barrel {enter} the room.
= John {make} the barrel + [A:John] [the barrel] {enter} the room.
John is putting the barrel in the room.

John {make} the barrel {roll} {enter} the room.
= John {make} the barrel + [A:John] [the barrel] {roll} + [A:John] [the barrel] {enter} the room.
John is making the barrel roll into the room.


The {make-roll} derivation which is of type (A, P) has the same meaning as {make} ... {roll} periphrasis.

John {make-roll} the barrel.
John is rolling the barrel.

John {make-roll} the barrel {enter} the room.
John is rolling the barrel into the room.


The agent-oriented (A, P) generic {make} by itself signifies that the agent is doing something to the patient, but does not imply what the result is, or even if there was a result.  When (P) and (P, T) predicates are made causative using {make}, (either by derivation or periphrasis) we _do_ know the result.  Other (A, P) predicates such as {push} have the same causative effect as {make}, but give more information about what the _agent_ is doing. 

John {make} the barrel.
John is doing something to the barrel.
[action vague, result unknown]

John {push} the barrel.
John is pushing on the barrel.
[action known, result unknown]

John {make} the barrel {enter} the room.
John is putting the barrel into the room.
[action vague, result known]

John {push} the barrel {enter} the room.
John is pushing the barrel into the room.
[action known, result known]


Serial predicates can be combined in any way that makes sense, but the combination is always considered a single event, not a "serial" one.

John {push} the barrel {roll}.
= John {push} the barrel + [A:John] [the barrel] {roll}.
John is pushing the barrel, rolling it.
NOT *John pushes the barrel which then rolls.
NOT *John pushes the barrel then rolls it.


In a complex event, it is not implied that one part of the event causes the other, though the possibility is left open (as in rolling the barrel and damaging it at the same time).  Part of the reason for this rule is allowing the freedom to reorder predicates without changing the meaning of the serial predicate.

John {make-roll} the barrel {damaged}.
John is rolling the barrel, damaging it.

John {make-damaged} the barrel {roll}.
John is damaging the barrel, rolling it.


One other basic predicate type, (A, T), needs to be mentioned.  The agent-oriented (A, T) generic {use} serves as an instrumental case. 

John {use} a forklift.
John is using a forklift.
= John is doing something with a forklist.


As already mentioned, when any (A, P) predicate is introduced into the clause, every patient-oriented predicate (P) and (P, T) acquires an agent slot semantically, but not syntactically.  Likewise, every predicate of type (A, T) acquires a *patient* slot semantically, but not syntactically.  The short rule is that (A, P) implies that all other predicates have both semantic agents and patients.

John {pushing} the barrel {use} forklift.
= John {push} barrel + [P:the barrel] John {use} a forklift.
John is pushing the barrel using a forklist.

John {use} a forklift {make} the barrel.
= [P:the barrel] John {use} a forklift + John {make} barrel.
John is using a forklift on the barrel.

John {push} the barrel {roll} {enter} the room {use} a forklift.
John is pushing the barrel, making it roll into the room, using a forklift.


Predicates of type (A, T) cannot be directly combined with patient-oriented predicates (P) and (P, T) -- it doesn't make sense to combine an instrumental-type predicate with a patient-oriented state because by themselves there are no common argument slots (the themes don't count).  However it's easy enough to add {make} periphrasis or derivation to permit an instrument.

*John {use} forklift the barrel {roll}.

*the barrel {roll} John {use} forklift.

John {make} the barrel {roll} {use} a forklift.
John is making the barrel roll using a forklift.


There is also a shortcut derivation to create an (A, P) predicate out of an instrument;  for example {use-forklift}.

John {use-forklift} the barrel {roll}.
John forklifts the barrel, rolling it.

John {use-forklift} the barrel {enter} the room.
John forklifts the barrel into the room.


Regarding the types of open class basic predicates -- predicates theoretically could have exactly zero or one place for each of the three argument types: Agent, Patient and Theme.  The unmarked argument order for OCBPs is constrained by the salience hierarchy A > P > T.   This gives us eight possible combinations (A, P, T), (A, P), (A, T), (A), (P, T), (P), (T), and (). The ones that seem to be needed are (P), (P, T), (A, P), and (A, T). 

Type (P) OCBPs represent generally static one-place predicates:  P is a duck, P is green, P is tall, P is broken, P is a prime number, etc.  (P) also represents a few involuntary actions like P shakes, P rolls, P sneezes, etc.  (P, T) OCBPs represent generally static two-place predicates:  P is the owner of T, P is the mother of T, P is the square root of  T, P is west of T, P has the shape of T, etc.  In this class per Morneau also go passive psychological experiencer predicates: P sees T, P loves T, P knows T, etc.

Types (P) and (P, T) OCBPs are inherently patient-oriented, where the exact thematic relationship of the "patient" P may range among what linguists call the patient, the subject, (confusingly) the theme, recipient, the experiencer, and a couple others.  In any case, a patient-oriented predicate always says something about what P is, referencing the theme T as needed.

Type (A, P) OCBPs like {push}, {hit} and {tell} describe us what the agent is doing.  They do not indicate any result or imply or effect on the patient.  They contrast with those derived from (P) like {make-roll}, {make-blue}, {make-broken}, which do describe the effect on the Patient and do not indicate how the agent accomplishing the result described.

The OCBP valencies that seem not needed are as follows. Triadic (A, P, T)s can be eliminated by decomposition into (A, P) + (A, T) dyads as shown above.

Monadic (A)s don't seem to be needed; ostensible actions that have no object whether P or T are suspect.  Some (A) candidates are plainly (P)s in disguise and are treated as such.  From these we can derive (A, P) transitives and (AP)-reflexives.  So from basic {shake} "The house shook" we can get {make-shake} "The storm shook the house" and {self-shake} "I shook off the cold."

Other (A) candidates describe actions complicated enough to be considered inherently voluntary, e.g. {walk} and {swim}.  In Morneau's system these are (AP)-reflexives, where the A is directing himself (P) to do something, but I put them with (P) in order to avoid creating a 5th class of OCBPs.  If someone wishes to use {self-walk} to emphasize agency, that's fine.  Either way, basic monadic (A)s and (AP)s disappear the system.

Any need for (T) can be eliminated because T's only real purpose is to express a relationship to P or A.  Ostensible (T) is really the passive form of (A, T) and (P, T).

I don't know of any real basic zero-arity () candidates, nor do they make sense to me.  Environmental predicates can be translated in the manner of "Rain falls [onto the fields]".