Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel-eecis!news.mathworks.com!newsfeed.internetmci.com!torn!blaze.trentu.ca!IVORY.trentu.ca!LJONES
From: ljones@ivory.trentu.ca (LEN JONES)
Subject: Searching for Direction in Semantics/Syntax Interaction
X-Nntp-Posting-Host: trentu.ca
Message-ID: <DvuGAt.Fv7@blaze.trentu.ca>
Sender: news@blaze.trentu.ca (USENET News System)
Reply-To: ljones@ivory.trentu.ca
Organization: Trent University, Peterborough
Date: Fri, 9 Aug 1996 00:06:29 GMT
Lines: 143

On Tue, 6 Aug 96  Steven Bayne @delphi.com wrote:

> For some time I have been investigating elementary semantics/syntax 
> interactions, ... paying close attention to cognitive features 
> associated with verbs in the hope of capturing some generalization
> of theoretical interest on semantics/syntax interaction ...
> Recently I have done some work on predicting interaction of the 
> syntactic categories of verb and preposition in sentences expressing 
> simple locomotion.

I have also been paying attention to locomotive verbs, primarily because 
they are the largest  category of verbs in English. Interestingly, most 
of these verbs incorporate some form of manner into their definition: 

   walk, stroll, amble, pace, stride, step, tramp, march, etc.   

Like all sentences, a sentence containing a locomotive verb has a finite 
limit on the number of different constituents of the sentence. Often these 
can be identified by a preposition.

The central arguments are: 
0) the verb,
1) the _agent_ that initiates the action,
2) the _patient_ that gets moved                (which may be the same as
                                                 the agent if intransitive),
3) the _vehicle_ which often looks like the patient, (tag 'on', 'in')
4) the _origin_ or source,                           (tag 'from','of','off')
5) the _destination_ or goal,                        (tag 'to', 'in', 'on') 
                which may not be reached (partitive - tag 'at', toward'),
6) a _path_ or intermediate location,                (tag 'via' or 'by')

other phrases are: 
9) an _instrument_, which often looks like the vehicle (tag 'with')
10) the _means_, an abstract instrument,               (tag 'by')  
10) the _manner_ of performance,                       (tag 'as')
11) a _commitive_ participant,                         (tag 'with')

and external to the central core:
11) Time of the event.
12) Location of the event, which may be an abstract domain.

   (Please feel free to disagree with my inclusion of vehicle.
    I find it a convenient notation)

Not all constituents appear in any one sentence for several reasons.
1) The information is not known.
2) The information is already common to speaker and listener.
3) The human brain has a limited capacity. Miller's 7 plus or minus 2 rule. 

I will state my views on sentence analysis.
Analyzing a sentence and deciding what the constituents are, if done at a
cognitive level, leads no where. Determining the rules at a precision that 
allows a computer program can analyze a series of sentences and provide a 
semantic analysis is more fruitful exercise. If you cannot provide such a 
level of detail then you haven't fully mastered the subject yet. 

However, appreciation of some sentences needs a real world knowledge.

   The hero kicked the little child.

A complete knowledge of all the linguistic rules will not capture the 
anomalous nature of a hero kicking a little child.
  

Syntax is a method of marking the constituents of a sentence so that humans 
can extract the speakers semantic intent. Obviously the computer program will 
generally have to follow the same procedures. English uses a mixture of word 
order and prepositional marking. The rules for this marking though 
subconsciously known and used by most native speakers are complex. 

Some speakers with various forms of aphasia cannot use these rules. They have 
to use real world knowledge to extract meaning. This often leads to problems 
given the sentence: 

   The dog was bitten by a man.

they can only interpreted it as:

   The dog bit a man.

There is much evidence from psycho-linguists that humans do use real world 
knowledge in speeding the analysis of speech, also, confirmation that 
grammatically ill-formed sentences take longer to process.

For normal speakers the rules are complicated by: 
1) the animate ordering hierarchy:

   me > you > them > other people > animals > living things > inanimate things

   (which is why we usually say 'fish and chips', almost never 'chips and fish')
 
2) possibly a constituent ordering hierarchy, which in part may be:
   
   volitional-agent > volitional-recipient > non volitional theme

    (thus we get 'John gave Mary a book')

these hierarchies introduce difficulties in sentences such as:

   The garden danced with fireflies.
   The streets of Paris ran with blood.

where the lack of suitable volitional agents cause the location of the event 
to be promoted to subject position and the non volitional agent to be marked 
as commitive. (You may disagree with this interpretation).  

3) differing function of prepositions synchronically and diachronically. 
I.E. usage varies by dialect and era.

There seems to be general rules, and rules which operate in a more restricted 
domain. For instance; some constituents drop the oblique marking in under 
certain conditions.

   He went from the sublime to the ridiculous very quickly. (fully marked)
   I went home.                     (unmarked)
   I came home.                     (unmarked)
   She came here yesterday.         (unmarked - movement relative to 'here')
   He went there on Friday.         (unmarked - movement relative to 'there') 
   He came from Seattle.                     (have to specify 'from' or 'to')
   Big John left Seattle in the year of '92. (unmarked 'leave' = 'go + from')

I haven't worked though them all yet but it appears conditioned entirely by 
the binary relation between verb and the destination/origin phrase.  

Partitive constructions are more controversial

   Hero swam the Hellespont        (completive - change of location)
   Hero swam across the Hellespont (may not have completely crossed it) 
   Eve threw the apple to Adam     (completive - change of possession & loc)
   Eve threw the apple at Adam     (may not have hit him) 
   Eve threw Adam the apple        (dative shift only if completive) 
  *Eve threw at Adam the apple     (not if partive)
 
I don't believe phrase structure rules, or predicate logic are powerful enough 
to handle all the rules and data involved in analyzing a natural language. 
Relations seem able to capture all the data about entities and their 
relationships, relational algebra and relational calculus then become the 
natural tools for manipulating this data.

Comments and enlightenment are welcome.

len jones

