Evolution of Syntactic Theory: From Psychology to Cognition

 
Syntax and cognition
 
Dick Hudson
Freie Universität Berlin, October 2015
1
Plan
 
1.
How syntactic theory has been influenced by psychology
2.
Why cognition?
3.
Phrase structure and dependency structure
4.
How to choose between PS and DS?
5.
A challenge for DS, and a cognitive solution
6.
Towards a new kind of DS
7.
New-DS and PS: are they notational variants?
8.
Conclusions
 
 
2
Ein redlich denkender Mensch verschmäht die Täuschung
 
1 The roots of phrase structure
a sincerely thinking person
scorns deception
a person
thinks sincerely
deception
is scorned
thought
is sincere
 
Who offered this analysis?
 
Wundt, Leipzig, 1900
 
A sincerely thinking person scorns deception
Gesammtvorstellung
3
 
So what?
 
Bloomfield took this analysis from Wundt
who thought top-down analysis was psychologically real
but was looking at the meaning, not the words
and turned it into his immediate-constituent analysis
which Chomsky turned into his phrase-structure grammar
using ideas from mathematics which used brackets
which he turned into trees without crossing branches.
So phrase structure is already based on assumptions about cognition.
4
 
2 Can we get away from cognition?
 
We can certainly try.
E.g. Integrational linguistics
But why would we want to try?
After all, language is surely a kind of knowledge.
So sooner or later our theories must meet theories about knowledge.
Jackendoff’s “graceful integration” of language with the rest of cognition
So we should at least try to build elementary ideas about other areas
of cognition into our theories of language.
5
 
Cognitive linguistics
 
This is one of the goals of cognitive linguistics
Including
Cognitive Grammar (Langacker)
Construction Grammar (Goldberg, Croft)
Word Grammar
My talk is about Word Grammar
developing since 1984
still changing
so some of this talk is new
6
 
3 Phrase structure or dependency structure?
 
Two traditions in syntax:
Phrase structure
Born in the USA (but inspired by Germany)
1933 Bloomfield
1957 Chomsky etc
Dependency structure
Much older
Born in the Middle East and Europe
But taught in the USA in the 19
th
 century (Reed and Kellogg diagrams)
1959 Tesnière
7
 
Phrase structure
 
The only relation recognised is the part-whole relation
Small babies cry.
small babies
cry.
small
babies
8
 
Dependency structure
 
The only relation recognised is the dependency between two words.
cry.
small
babies
stemma
cry.
small
babies
WG
adjunct
subject
9
 
How a DS grammar works
 
Every word has a valency
the dependents that it needs
(WG only) also its need for a ‘parent’ (a word on which it depends)
These needs must be satisfied by other words
Totally ‘bottom-up’.
Every word also has a meaning
lexical meaning
modified by dependents
babies 
 means ‘small babies’ when modified by 
small
cry 
means ‘small babies cry’ when modified by 
babies
 modified by 
small
10
 
4 How to choose between PS and DS?
 
Consider the facts
e.g. maybe c-command is important and requires PS?
Look for elegance
count the nodes
Look at general cognition
what kinds of relations can we recognise in general?
answer: many different kinds!!!
part-whole relations
social relations among individuals
spatial relations among objects
relations between events and their participants etc etc etc
11
 
So the winner is ...
 
Neither PS nor DS
because they both recognise only one kind of relation
and we know that our minds can recognise many different kinds.
But DS is better than PS
because the evidence for word-word relations is stronger than for phrases:
lexical selection, e.g. DEPEND + ON
idioms, e.g. TAKE + CARE
government, e.g. MIT + Dative
But there is a little evidence for phrase-like units ...
12
 
5 A challenge for DS
 
a big French house 
= a house which is big and French
But: 
a typical French house 
= a house which is typical of French
houses.
Noticed by Oesten Dahl 1980
typical
French
house
French house
typical French house
typical
French
house
W
Where is ‘French
house’?
13
 
Solution: use what general cognition offers
 
Knowledge is a network of atomic nodes.
The network distinguishes different kinds of relation
including a vast and open collection of ‘relational concepts’, created as
needed
The ‘isa’ relation allows default inheritance.
We create temporary nodes for experiences.
e.g. someone I saw on the street, ‘personX’
14
 
A tiny cognitive network
Berlin
city
capital
city
Germany
Berliner
FU
capital of
citizen of
university
citizen of
‘isa’
personX
new
 
node
15
 
Default inheritance
 
This is part of node-creation.
If A isa B, then the properties of A always override those of B.
So we can make generalisations even when there are exceptions.
16
 
Default inheritance in birds
2
leg
wing
bird
beak
1
eggs
flying
babies
move
-ment
part
part
part
#
#
#
1
#
penguin
flying
move
-ment
0
#
17
NB | ‘with’, not + ‘and’. Statistics: p(x|y)
means ‘the probability of x in the context
of y’
 
Node-creation
 
We create a new node (‘X’) for ongoing experience so that
we can enrich it by classifying it as a Y, and inheriting from Y
we can distinguish the experience from Y
we can accommodate irregularities.
And we continue to enrich it in the light of new information (Z)
by creating a further node (‘X|Z’, ‘X with Z’)
with isa links to X.
These nodes allow us to remember earlier states
and they’re the material of detailed ‘constructions’.
For example ...
X
Y
a
r
b
r
X|Z
18
concept
|
 
What is it?
tin
paint brush
for
paint tin
?
contains
paint
contains
19
 
6 Towards a new kind of DS
 
Assume one initial node per word.
This inherits directly from some lexeme in the grammar.
e.g. in 
paint tin
, we create one concept for each word token:
 
But add an extra node for each dependent.
This shows how the word’s meaning is affected by each dependent.
e.g. we create an extra concept for ‘
tin 
as modified by 
paint’
:
 
 
And we link the two nodes by ‘isa’.
paint
tin
tin | paint
tin
20
tin
  |  
paint
 
New-DS:  
paint tin
TIN
PAINT
dependent
TIN | PAINT
tin
‘tin’
means
paint
‘tin for
paint’
means
‘paint’
means
21
 
Back to 
typical French houses.
typical
French
house
typical
French
house
house | French
house|French | typical
house
 modified
by 
French
house
 | 
French 
modified by
typical
French
house
means
typical
French
house
means
house
means
22
 
New-DS and PS
typical
French
house
house | French
house|French | typical
typical
French
house
house + French
house+French + typical
Notational variants??
PS
new-DS
23
 
Isa, not part-of
 
A isa B: like ‘Mary isa linguist’
shared properties
same size
B is-part-of A: like ‘Mary’s foot is-part-of Mary’
different properties
different size
B (house)
A (house | French)
B (house)
A (house + French)
 
B is-part-of A
 
A isa B
24
 
No unary branching in new-DS.
 
PS needs both A and B because they have different properties, even
when they have the same size.
New-DS doesn’t and can’t.
‘higher’ nodes are only needed where there’s a dependent.
Hurry!/1
PS
new-DS
word
Hurry!/2
Hurry!/3
Hurry! 
(word)
VP
sentence
25
 
New-DS guarantees headedness
 
A problem for DS?
Student after student came in
. (Jackendoff)
What is the head?
Answer: the first 
student
,
 
just like 
tin of paint
.
But why no determiner?
Stipulated, as in 
to school, at home
wine from France 
but: 
the wine of France
A construction definable, as usual, in terms of dependencies
26
 
In new-DS single dependencies are
constructions
 
e.g. GIVE NP A HARD TIME
27
GIVE
direct
verb
subject
indirect
GIVE|time
direct
A|HARD TIME
 
New-DS allows meaning-order mismatches
 
People are sometimes very tall
.
= Some people are very tall.
= sometimes (people are very tall)
John is typically late again.
= 
John is late again, and John being late is typical.
I needed a small brass screw, but I could only find a steel one.
one 
 = ‘small screw’, not ‘brass screw’ or ‘small brass screw’!
28
 
How does New-DS allow this mismatch?
 
I needed a small brass screw, but I could only find a steel one.
one 
 = ‘small screw’.
small
brass
screw
screw | brass
screw|brass & small
screw | small
steel
one
‘small screw’
sense
sense
?
29
 
8 Conclusions
 
Syntactic theory should build on cognitive science.
We should assume that our minds can apply any general-purpose
cognitive machinery to language.
This affects our assumptions about syntactic structure.
It throws new light on the old dispute about PS versus DS.
It allows us to develop a new version of DS which is more similar to
PS.
But even new-DS is different from PS
and better!
30
 
Danke für Ihre Aufmerksamkeit und Geduld!
 
This slide show is available at
dickhudson.com/talks
Word Grammar offers much, much more ...
see dickhudson.com/word-grammar/
31
Slide Note

Embed
Share

Syntactic theory has evolved significantly, influenced by psychology and cognition. The debate between phrase structure and dependency structure continues, with a shift towards integrating cognitive principles. Cognitive linguistics, including Cognitive Grammar and Construction Grammar, play a key role in this evolution. The discussion delves into the historical roots, challenges, and the ongoing development of Word Grammar since 1984.


Uploaded on Sep 27, 2024 | 0 Views


Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

E N D

Presentation Transcript


  1. Syntax and cognition Dick Hudson Freie Universit t Berlin, October 2015 1

  2. Plan 1. How syntactic theory has been influenced by psychology 2. Why cognition? 3. Phrase structure and dependency structure 4. How to choose between PS and DS? 5. A challenge for DS, and a cognitive solution 6. Towards a new kind of DS 7. New-DS and PS: are they notational variants? 8. Conclusions 2

  3. Gesammtvorstellung 1 The roots of phrase structure a sincerely thinking person scorns deception a person thinks sincerely deception is scorned thought is sincere Who offered this analysis? Ein redlich denkender Mensch verschm ht die T uschung A sincerely thinking person scorns deception Wundt, Leipzig, 1900 3

  4. So what? Bloomfield took this analysis from Wundt who thought top-down analysis was psychologically real but was looking at the meaning, not the words and turned it into his immediate-constituent analysis which Chomsky turned into his phrase-structure grammar using ideas from mathematics which used brackets which he turned into trees without crossing branches. So phrase structure is already based on assumptions about cognition. 4

  5. 2 Can we get away from cognition? We can certainly try. E.g. Integrational linguistics But why would we want to try? After all, language is surely a kind of knowledge. So sooner or later our theories must meet theories about knowledge. Jackendoff s graceful integration of language with the rest of cognition So we should at least try to build elementary ideas about other areas of cognition into our theories of language. 5

  6. Cognitive linguistics This is one of the goals of cognitive linguistics Including Cognitive Grammar (Langacker) Construction Grammar (Goldberg, Croft) Word Grammar My talk is about Word Grammar developing since 1984 still changing so some of this talk is new 6

  7. 3 Phrase structure or dependency structure? Two traditions in syntax: Phrase structure Born in the USA (but inspired by Germany) 1933 Bloomfield 1957 Chomsky etc Dependency structure Much older Born in the Middle East and Europe But taught in the USA in the 19th century (Reed and Kellogg diagrams) 1959 Tesni re 7

  8. Phrase structure The only relation recognised is the part-whole relation Small babies cry. cry. small babies small babies 8

  9. Dependency structure The only relation recognised is the dependency between two words. WG stemma subject adjunct babies cry. babies cry. small small 9

  10. How a DS grammar works Every word has a valency the dependents that it needs (WG only) also its need for a parent (a word on which it depends) These needs must be satisfied by other words Totally bottom-up . Every word also has a meaning lexical meaning modified by dependents babies means small babies when modified by small cry means small babies cry when modified by babies modified by small 10

  11. 4 How to choose between PS and DS? Consider the facts e.g. maybe c-command is important and requires PS? Look for elegance count the nodes Look at general cognition what kinds of relations can we recognise in general? answer: many different kinds!!! part-whole relations social relations among individuals spatial relations among objects relations between events and their participants etc etc etc 11

  12. So the winner is ... Neither PS nor DS because they both recognise only one kind of relation and we know that our minds can recognise many different kinds. But DS is better than PS because the evidence for word-word relations is stronger than for phrases: lexical selection, e.g. DEPEND + ON idioms, e.g. TAKE + CARE government, e.g. MIT + Dative But there is a little evidence for phrase-like units ... 12

  13. 5 A challenge for DS a big French house = a house which is big and French But: a typical French house = a house which is typical of French houses. Noticed by Oesten Dahl 1980 typical French house WWhere is French house ? French house typical French house typical French house 13

  14. Solution: use what general cognition offers Knowledge is a network of atomic nodes. The network distinguishes different kinds of relation including a vast and open collection of relational concepts , created as needed The isa relation allows default inheritance. We create temporary nodes for experiences. e.g. someone I saw on the street, personX 14

  15. A tiny cognitive network city isa capital city new node citizen of capital of Germany Berliner Berlin university FU citizen of personX 15

  16. Default inheritance This is part of node-creation. If A isa B, then the properties of A always override those of B. So we can make generalisations even when there are exceptions. 16

  17. Default inheritance in birds 1 beak # part eggs babies bird leg part # move -ment # flying 2 1 # part wing penguin move -ment # flying 0 17

  18. Node-creation a Y r We create a new node ( X ) for ongoing experience so that we can enrich it by classifying it as a Y, and inheriting from Y we can distinguish the experience from Y we can accommodate irregularities. And we continue to enrich it in the light of new information (Z) by creating a further node ( X|Z , X with Z ) with isa links to X. These nodes allow us to remember earlier states and they re the material of detailed constructions . For example ... b X r X|Z NB | with , not + and . Statistics: p(x|y) means the probability of x in the context of y 18

  19. concept What is it? tin paint brush paint tin | for contains contains ? paint 19

  20. 6 Towards a new kind of DS Assume one initial node per word. This inherits directly from some lexeme in the grammar. e.g. in paint tin, we create one concept for each word token: paint tin But add an extra node for each dependent. This shows how the word s meaning is affected by each dependent. e.g. we create an extra concept for tin as modified by paint : tin | paint And we link the two nodes by isa . tin 20

  21. New-DS: paint tin PAINT TIN | PAINT TIN tin tin | paint paint dependent means means means tin for paint tin paint 21

  22. Back to typical French houses. house | French modified by typical typical French house typical French house house|French | typical house modified by French means French house house | French means typical French house house means 22

  23. New-DS and PS house|French | typical PS house | French house+French + typical typical French house house + French new-DS Notational variants?? typical French house 23

  24. Isa, not part-of A (house + French) A (house | French) A isa B B is-part-of A B (house) B (house) A isa B: like Mary isa linguist shared properties same size B is-part-of A: like Mary s foot is-part-of Mary different properties different size 24

  25. No unary branching in new-DS. PS needs both A and B because they have different properties, even when they have the same size. New-DS doesn t and can t. higher nodes are only needed where there s a dependent. sentence Hurry!/3 PS new-DS Hurry!/2 VP Hurry! (word) Hurry!/1 word 25

  26. New-DS guarantees headedness A problem for DS? Student after student came in. (Jackendoff) What is the head? Answer: the first student,just like tin of paint. But why no determiner? Stipulated, as in to school, at home wine from France but: the wine of France A construction definable, as usual, in terms of dependencies 26

  27. In new-DS single dependencies are constructions e.g. GIVE NP A HARD TIME subject verb direct GIVE indirect direct A|HARD TIME GIVE|time 27

  28. New-DS allows meaning-order mismatches People are sometimes very tall. = Some people are very tall. = sometimes (people are very tall) John is typically late again. = John is late again, and John being late is typical. I needed a small brass screw, but I could only find a steel one. one = small screw , not brass screw or small brass screw ! 28

  29. How does New-DS allow this mismatch? I needed a small brass screw, but I could only find a steel one. one = small screw . screw|brass & small ? small screw sense sense screw | small screw | brass steel one small brass screw 29

  30. 8 Conclusions Syntactic theory should build on cognitive science. We should assume that our minds can apply any general-purpose cognitive machinery to language. This affects our assumptions about syntactic structure. It throws new light on the old dispute about PS versus DS. It allows us to develop a new version of DS which is more similar to PS. But even new-DS is different from PS and better! 30

  31. Danke fr Ihre Aufmerksamkeit und Geduld! This slide show is available at dickhudson.com/talks Word Grammar offers much, much more ... see dickhudson.com/word-grammar/ 31

Related


More Related Content

giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#