CML 2010 DUBROVNIK THE ARGUMENT-BASED COMPUTATION: SOLVING THE BINDING PROBLEM Velina Slavova, NBU...
-
date post
16-Jan-2016 -
Category
Documents
-
view
216 -
download
0
Transcript of CML 2010 DUBROVNIK THE ARGUMENT-BASED COMPUTATION: SOLVING THE BINDING PROBLEM Velina Slavova, NBU...
CML 2010 DUBROVNIK
THE ARGUMENT-BASED COMPUTATION: SOLVING THE BINDING PROBLEM
Velina Slavova, NBU
Alona Soschen, MIT
CML 2010 DUBROVNIK
We offer an experimental support of the developed formal model of Argument Based Syntax.
In this paper, we further developed the argument-based model of syntactic operations that is argued to represent the key to basic mental representations.
The binding problem arises when the same noun occupies either a Subject or an
Object position.
The cat chases the mouse
The mouse chases the cat
The results of our experiments confirmed that semantic relations between the images in conceptual nets influence syntactic computation.
We use the absence of of Accusative Case marker in Bulgarian and its rather free word order, and we show how the syntactic structure of ambiguous sentences is interpreted by means of other cognitive mechanisms.
CML 2010 DUBROVNIK
by the (formal) semantic system (Hauser et al., 2002).
Introduction
LanguageSemantic
system
Following one of the widely accepted linguistic theories, the key component of Faculty of Language is a computational system (narrow syntax)
that generates internal representations
and maps them into the conceptual-intentional interface
CML 2010 DUBROVNIK
Language
SYNTACTIC RULES ?
Semantic knowledge
COMPUTATION
There is a consensus that the core property of FL is recursion, which is attributed to narrow syntax.
The claim is that this computation is based on a primitive operation that takes already constructed objects to create a new object. This basic operation, called
“Merge”, provides a “language of thought”, an internal system to allow preexistent conceptual resources to construct expressions (Chomsky, 2006).
Introduction
CML 2010 DUBROVNIK
One step in this direction was provided in Slavova and Soschen (2007). Syntactic structures, presented in the traditional sense of Chomskyan theory (Bare Phrase Structures, XP-structures), were re-defined in terms of finite recursive binary trees.
XP
XP
XXP
X’
S : The boy likes the girl
The boy/ NPV’/ likes the girl
NP/ The girlSpecifier
Head X Complement
The study of syntactic recursion by mathematical means may provide valuable insights into the principles underlying the human language.
Introduction
CML 2010 DUBROVNIK
X’XP
XXPX’ XP
X XP XPX’ XPX’
XPX’ X XPX’ X XP X’
XP XP
X XP XP X’
X XPX’ XXPX’
XPX’ XPXP X’
XP
X’XP
XXPX’ XP
X XP XPX’ XPX’
XPX’ X XPX’ X XP X’
XP XP
X XP XP X’
X XPX’ XXPX’
XPX’ XPXP X’
XP
X’XP
XXPX’ XP
X
XP XPX’XP
X’
XPX’ X XPX’ X XP X’
XP XP
XXP XP X
’X XPX’ XXPX
’
XPX’ XPXP X’
XP
X’XP
XXPX’XP
X
XP XPX’ XPX’
XPX’X
XPX’ X XP X’
XP XP
X XP XP X’
X XPX’ XXPX’
XPX’ XPXPX’
XP
X
X
X X
X X X
X’XP
XPX’ XP
XP XPX’ XPX’
XPX’ XPX’ XP X’
XP XP
XP XP X’
XPX’ XPX’
XPX’ XPXP X’
XP
XP
The “traditional syntactic tree” was modified:
the nodes related to syntactic role of verbs were discarded.
Introduction
CML 2010 DUBROVNIK
X
X
X
X X
X X X
X’XP
XPX’ XP
XP XPX’ XPX’
XPX’ XPX’ XP X’
XP XP
XP XP X’
XPX’ XPX’
XPX’ XPXP X’
XP
XP
One reason to introduce this modification was to build a structure that complies with the principles of optimization, namely with the principle of efficient growth (Soschen 2006, 2008).
Introduction
CML 2010 DUBROVNIK
X
X
X X
X X X
X’XP
XPX’ XP
XP XPX’ XPX’
XPX’ XPX’ XP X’
XP XP
XP XP X’
XPX’ XPX’
XPX’ XPXP X’
XP
XP
CML 2010 DUBROVNIK
XP
X’XP
XXPX’ XP
X XP XPX’ XPX’
XPX’ X XPX’ X XPX’
XP XP
X XP XPX’
X XPX’ X XPX’
XPX’ XPXPX’
XP
The structure obtained in this way is a tree of Fibonacci Introduction
CML 2010 DUBROVNIK
XP
XP
XP XP
XP XP XP
XP XP XPXP XP
XP XP XP XPXP XPXP XP
Introduction
CML 2010 DUBROVNIK
XP
XP XP
X3
Ø
Ø
XPXP XPXP
XPXP XPXP
X4
Ø
Ø
X5
Ø
XP
X2
ØX1
Ø Ø
Introduction
CML 2010 DUBROVNIK
XP
XP XP
X3
Ø
Ø -merge
Ø -merge
ØØ -merge
Ø -merge
XPXP XPXP
XPXP XPXP
X4
Ø
Ø
Ø -merge
X5
Ø
Ø -merge
XP
X2
Ø
Ø -merge
X1
ØØ -merge
Ø
merge merge
merge
merge
Language of thought?
This tree can be seen as is an operator – it “performs” a bottom-up Merge, its nodes are the results of Merge. The tree shows the patterns of relating arguments Xs.
CML 2010 DUBROVNIK
This maximal configuration corresponds to thematic roles agent, recipient, and theme (the arguments in a sentence).
These schemes represent all possible configurations and relations between arguments in the human theta-role Semantic Space.
Introduction
CML 2010 DUBROVNIK
The difficulty of designing an appropriate experiment is that mental computation runs on a deep (pre-linguistic) level and cannot be captured on the lexical level by a standard experiment.
EXPERIMENT 1
One possible way to extract some information about the primary mechanisms is to force the mental system to solve ambiguities on the lexical level and to analyze the system’s response.
Language
Rules,Operators
Semantic knowledge
“computation”
Introduction
CML 2010 DUBROVNIK
EXPERIMENT 1
In Bulgarian, sentences of type:
are ambiguous where the used verb allows a Recipient (to whom).
Subject Verb the Object на the Z
NB Bulgarian exhibits certain peculiarities that set it apart from other Slavic languages, such as elimination of Case marking and the development of a suffixed definite article.
They assign two different meanings to Z - Recipient and Possessor:
Subject Verb the Object to the Z
Subject Verb the Object of the Z
Introduction
CML 2010 DUBROVNIK
Test statement
Анна подаде стола на директора.
Анна продаде ябълките на момчето.
Елена продаде къщата на кучето.
Иван показа пътеката на баща си.
Иван продаде къщата на баща си.
Кумчо Вълчо продаде къщата на кучето.
Мария показа колата на съседката.
Мария продаде колата на съседката.
Михаил продаде къщата на съседа си.
Монтьорът показа колата на съседката.
Петър донесе стола на директора.
Петър показа къщата на баща си.
Тапицерът донесе стола на директора.
In normal listening or reading-comprehension conditions, native Bulgarian speakers are never mistaken about the conveyed meaning - they interpret one of these two senses depending on the context.
EXPERIMENT 1
Introduction
CML 2010 DUBROVNIK
The subjects of our experiment wore 62 students with different backgrounds (economists, sociologists, biologists, linguists, engineers etc.) all of them fluent speakers of French
Test statement
Анна подаде стола на директора.
Анна продаде ябълките на момчето.
Елена продаде къщата на кучето.
Иван показа пътеката на баща си.
Иван продаде къщата на баща си.
Кумчо Вълчо продаде къщата на кучето.
Мария показа колата на съседката.
Мария продаде колата на съседката.
Михаил продаде къщата на съседа си.
Монтьорът показа колата на съседката.
Петър донесе стола на директора.
Петър показа къщата на баща си.
Тапицерът донесе стола на директора.
Translate into French!
(Quickly, please)
The 124 written translations of the test statements were stored in a database.
EXPERIMENT 1
In order to understand which meaning is captured first by the subjects in our experiment, we used the fact that the sentence structure, including word order, is exactly the same in French.
Introduction
CML 2010 DUBROVNIK
Test statement
Анна подаде стола на директора.
Анна продаде ябълките на момчето.
Елена продаде къщата на кучето.
Иван показа пътеката на баща си.
Иван продаде къщата на баща си.
Кумчо Вълчо продаде къщата на кучето.
Мария показа колата на съседката.
Мария продаде колата на съседката.
Михаил продаде къщата на съседа си.
Монтьорът показа колата на съседката.
Петър донесе стола на директора.
Петър показа къщата на баща си.
Тапицерът донесе стола на директора.
Translate into French!
(Quickly, please)
EXPERIMENT 1
The crucial difference is that the preposition на is translated in French as
“à” (to) for the Recipient-meaning and as “de” (of) for the Possessor-meaning.
Introduction
CML 2010 DUBROVNIK
Here the result:
The important observation is that these changes do not depend on the verb.
Introduction
CML 2010 DUBROVNIK
The Z The Y Verb X на
Concept Achievement
Concept Preposition
Concept
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
CML 2010 DUBROVNIK
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
Semantic Merge (M) is modeled as binary operation, performed in sequential progression between concepts
verbY*X* Z*
The result of M consists of a couple, in which each element obtains an image , representing the concept in the semantic context of the other member of the couple.
For example, the image of the concept X* within the couple [X*V] is the image of X* as Subject, performing V.
verbX* X Acts V
CML 2010 DUBROVNIK
Z*= Recipient ( X*)
X* =
M (X*,V) = [X*,V]
X* = X*V (Y*, ?R)
X* V (?O, ?R) Y* Z*
M ([X*,Y* ], [X*,Z*])
X*= X*V (Y*, Z*)
Y* = Object of X* to Z*Z*= Recipient of X* Acts Y*
1.
4.
на
M (X*, Y*) = [X*, Y*] 3.
= X*V (?O, ?R)
Y* = Object X* V
M (X*, Z*) = [ X*, Z*] 2.
SEMANTIC MERGE – experimental results 1
Recipient scheme
PRIMARY
CML 2010 DUBROVNIK
X* =
M (X*,V*) = [X*,V*]
X* = X* Acts (?what, to Ø)
X* V*( ?O, ?R) Y* Z*
M ([X*, Y* ], [X*, Ø ])
X*= X* Acts Y* of Z*Y* = belongs to Z* and Acted by X* Z*= Possessor of Y*
1.
4.
на
M (X*, Ø) = [X*, Ø]2b.
= X* Acts (?O, ?R)
M (X*, Y*) = [ X*, Y*] 3.
Z* = (Z*Y*) = Poss of Y*
Y* = (Z*Y*) = belongs to Z*
Ø2a.
Y* = X* Acts Y*, to Ø)
= X* Acts (?O)
Possessor scheme
IF RECIPIENT IS REJECTED
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
Z*= Recipient ( X*)
X* =
M (X*,V) = [X*,V]
X* = X*V (Y*, ?R)
X* V (?O, ?R) Y* Z*
M ([X*,Y* ], [X*,Z*])
X*= X*V (Y*, Z*)
Y* = Object of X* to Z*Z*= Recipient of X* Acts Y*
1.
4.
на
M (X*, Y*) = [X*, Y*] 3.
= X*V (?O, ?R)
Y* = Object X* V
M (X*, Z*) = [ X*, Z*] 2.
Recipient scheme
The experimental results show that if in the semantic space there exists a previous knowledge about Z* as a natural Possessor of Y*, that influences the process on step 2.
Z Y Previous semantic knowledge
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
X* [X*, V] =
M (X*,V) = [X*,V]
X* V (?O, ?R) Y* Z*
1.
на
= X*V (?O, ?R)
The experimental results show that if in the semantic space there exists a previous knowledge about Z* as a natural Possessor of Y*, that influences the process on step 2.
Z Y
Z* = (Z*Y*) = Poss of Y*
Y* = (Z*Y*) = belongs to Z*
2a.
X* = X* Acts (?what, to Ø)
M ([X*, Y* ], [X*, Ø ])
X*= X* Acts Y* of Z*Y* = belongs to Z* and Acted by X* Z*= Possessor of Y*
4.
M (X*, Ø) = [X*, Ø]2b.
M (X*, Y*) = [ X*, Y*] 3. Y* = X* Acts Y*, to Ø)
= X* Acts (?O)
Possessor scheme
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
Z*= Recipient ( X*)
X* =
M (X*,V) = [X*,V]
X* = X*V (Y*, ?R)
X* V (?O, ?R) Y* Z*
M ([X*,Y* ], [X*,Z*])
X*= X*V (Y*, Z*)
Y* = Object of X* to Z*Z*= Recipient of X* Acts Y*
1.
4.
на
M (X*, Y*) = [X*, Y*] 3.
= X*V (?O, ?R)
Y* = Object X* V
M (X*, Z*) = [ X*, Z*] 2.
X* Y* на
Recipient scheme
Y X
The knowledge about Y* as usual Object of actions of X* forms in the semantic space a relation X*-Y*.
Previous knowledge
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
Z*= Recipient ( X*)
X* =
M (X*,V) = [X*,V]
X* = X*V (Y*, ?R)
X* V (?O, ?R) Y* Z*
1.
на
M (X*, Y*) = [X*, Y*] 3.
= X*V (?O, ?R)
Y* = Object X* V
M (X*, Z*) = [ X*, Z*] 2.
X* Y* на
Recipient scheme
Y X
The knowledge about Y* as usual Object of actions of X* forms in the semantic space a relation X*-Y*.
M ([X*, Y* ], [X*, Ø ])4.
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
X* [X*, V] =
M (X*,V) = [X*,V]
X* V (?O, ?R) Y* Z*
1.
на
= X*V (?O, ?R)
Possessor scheme
The knowledge about Y* as usual Object of actions of X* forms in the semantic space a relation X*-Y*.
Y
Z* = (Z*Y*) = Poss of Y*
Y* = (Z*Y*) = belongs to Z*
2a.
X* = X* Acts (?what, to Ø)
M ([X*, Y* ], [X*, Ø ])
X*= X* Acts Y* of Z*Y* = belongs to Z* and Acted by X* Z*= Possessor of Y*
4.
M (X*, Ø) = [X*, Ø]2b.
M (X*, Y*) = [ X*, Y*] 3. Y* = X* Acts Y*, to Ø)
= X* Acts (?O)
X
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
Z*= Recipient ( X*)
X* =
M (X*,V) = [X*,V]
X* = X*V (Y*, ?R)
X* V (?O, ?R) Y* Z*
M ([X*,Y* ], [X*,Z*])
X*= X*V (Y*, Z*)
Y* = Object of X* to Z*Z*= Recipient of X* Acts Y*
1.
4.
на
M (X*, Y*) = [X*, Y*] 3.
= X*V (?O, ?R)
Y* = Object X* V
M (X*, Z*) = [ X*, Z*] 2.
X* Y* на
Recipient scheme
ZX
When between X* and Z* exists a strong semantic relation, Z* takes the (natural) role of Recipient of the actions of X*.
SEMANTIC MERGE – experimental results 1
CML 2010 DUBROVNIK
The general conclusion is that people very much construct the meaning of what has been said.
CML 2010 DUBROVNIK
(SVO): децата прочетоха писмoто.
‘The children have read the letter’.
(SOV): децата писмoто прочетоха.
‘The children the letter have read’.
(OSV): писмoто децата прочетоха.
‘The letter the children have read’.
(OVS): писмoто прочетоха децата.
‘The letter have read the children’.
(VOS): прочетоха писмoто децата.
‘Have read the letter the children’.
(VSO): прочетоха децата писмoто.
‘Have read the children the letter’.
EXPERIMENT 2
Subject
Verb
Object
(Bulgarian)
Although the Bulgarian nouns are not marked for Case, the word order is rather free.
Thus, in a Bulgarian the sentence ‘The children have read the letter’ can be expressed as the following:
the Noun 1 Verb the Noun 2
The children
have read
the letter
the Noun 1 Verb the Noun 2
CML 2010 DUBROVNIK
EXPERIMENT 2
Subject
Verb
Object
(Bulgarian)
The children
have read
the letter
the Noun 1 Verb the Noun 2
Although SVO is the basic WO, all permutations are possible; they are grammatically correct, even thought some are used mostly in poetry.
According to the Institute of Bulgarian at the Bulgarian Academy of Sciences, this grammatical particularity of permutation is possible because of the agreement between Subject and Verb which clarifies the role of Object.
But whatIf the Verb is in agreement with both nouns?
s
There exists in Bulgarian a particular grammatical operation (clitic doubling) that marks Object in cases of a reverse word order, as in:
(OVS): Иван го поздрави Мария.
Ivan him congratulated Maria
‘Maria congratulated Ivan’.
CML 2010 DUBROVNIK
EXPERIMENT 2
Subject
Verb
Object
(Bulgarian)
Noun 1
Noun 2
Noun 1 Verb Noun 2
The clitic (го/ги) is obligatory only when the subject and the object are in the third person, and they are both either singular or plural.
According to the Institute of Bulgarian at the Bulgarian Academy of Sciences, When the meaning is clear from the context, the clitic can be omitted.
(OVS): Иван го поздрави Мария.
Ivan him congratulated Maria
‘Maria congratulated Ivan’.
го/ги
(OVS): Ролите озвучиха артистите: (list).
The roles sound-screened the artists:
‘The artists (from the list) sound-screened the roles’.
for example:
CML 2010 DUBROVNIK
EXPERIMENT 2 (Merge patterns)
M (X*,V) = [X*,V]
X* = X*V (Y)
X* V (?O) Y*
1.
M (X*, Y*) = [X*, Y*] 2.
X Acts (?O)
Y* = Object X* V
Noun 1 Noun 2verb The first mental operator in this treatment is to merge nouns with V*
Stage I : Merge Subject-Verb
M (X*V*) = [X*, V] (?O)
X* = X* Acts (?O)
Y*
X*
Stage II : Merge Acting Subject with Object
Y* [ X*, Y*] = Object X* V
X*
X* V Object Y*
CML 2010 DUBROVNIK
EXPERIMENT 2 (Merge patterns)
M (Y*,V) = [Y*,V]
X* V (?O) Y*
1.
M (Y*, X*) = [Y*, X*] 2.
= Y Acts (?O)
Y* = X*V (Y)
X* = Object Y* V
Stage I : Merge Subject-Verb
M (Y*V*) = [Y*, V] (?O)
Y* = Y* Acts (?O)
Stage II : Merge Acting Subject with Object
X*
Y*
Y*
Y* V Object X*
CML 2010 DUBROVNIK
EXPERIMENT 2 (Merge patterns)
X*
M (X*,V) = [X*,V]
X* = X*V (Y)
X* V (?O) Y*
1.
M (X*, Y*) = [X*, Y*] 2.
= X*V (?O)
Y* = Object X* V
Y*
M (Y*,V) = [Y*,V]
Y* = Y*V (X)
X* V (?O) Y*
1.
M (Y*, X*) = [Y*, X*] 2.
= Y*V (?O)
X* = Object Y* V
We assume that these operations are semantically supported by the meaning of the concepts X* and Y*.
X* Y*
IF THERE ARE NOT GRAMMATICAL MARKERS…
CML 2010 DUBROVNIK
EXPERIMENT 2 (Bulgarian)
In this case, either X or Y can appear as Object:
We used Bulgarian sentences of type:
Where V is in agreement with both X and Y.
X(noun) VerbY(noun)
Subject
Verb
Object
X(noun)
Verb
Y(noun)
CML 2010 DUBROVNIK
EXPERIMENT 2 (Bulgarian)
The following folk verses that contain sentences of the above kind were used in our experiment:
CML 2010 DUBROVNIK
Живееше мишка, сива и красиваOnce upon a time there lived a mouse, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoМишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
CML 2010 DUBROVNIK
Живееше куче, сиво и красивоOnce upon a time there lived a dog, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoКучето Котанчо хвана и изчезна под дивана.The dog Kotantcho caught and disappeared under the sofa
CML 2010 DUBROVNIK
Живееше мишка, сива и красиваOnce upon a time there lived a mouse, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoМишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
CML 2010 DUBROVNIK
Живееше куче, сиво и красивоOnce upon a time there lived a dog, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoКучето Котанчо хвана и изчезна под дивана.The dog Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The two sentences in the verses have identical structure X* Y* V* and there are no grammatical markers indicating which noun is Subject and which is Object.
CML 2010 DUBROVNIK
The subjects of our experiment wore 36 students with different backgrounds (economists, sociologists, biologists, linguists, engineers etc.) all of them fluent speakers of French
Translate into French!
(Quickly, please)
The written translations of the test statements were stored in a database.
We designed our experiment as a translation (from Bulgarian to French) task.
In French the word order is fixed.
Thus, our subjects had to assign a fixed word order to their French translations, thus bringing out their interpretation of the same noun as either Subject or Object.
Живееше мишка, сива и красиваOnce upon a time there lived a mouse, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoМишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
Живееше куче, сиво и красивоOnce upon a time there lived a dog, grey and beautifulнейде на тавана, дето бе дивана. Somewhere in the attic, where the sofa wasПояви се тук Котанчо, котаракът на Стоянчо.There appeared Kotantcho, the cat of SoyantchoКучето Котанчо хвана и изчезна под дивана.The dog Kotantcho caught and disappeared under the sofa
Subject
Verb
Object
Noun 1
Noun 2
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Кучето Котанчо хвана и изчезна под дивана.The dog Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The dog
Caught(activity)
Kotantcho (cat)
SUBJECT Verb OBJECT
Result: 17 (100%)
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Мишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The mouse
Caught(activity)
Kotantcho (cat)
Result: 12 (100%)
‘retell the story in two sentences’ condition
SUBJECT Verb OBJECT All translations were structured as S V O (le chat a attrapé la sourris).
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Мишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The mouse
Caught(activity)
Kotantcho (cat)
Result: 16 (72,7%)
‘translate the story’ condition
SUBJECT Verb OBJECT
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Мишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The mouse
Caught(activity)
Kotantcho (cat)
Result: 4 (18,2%)
‘translate the story’ condition
SUBJECT Verb OBJECT
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Мишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The mouse
Caught(activity)
Kotantcho (cat)
NOT CLEAR Result: 2
‘translate the story’ condition
SUBJECT Verb OBJECT
The cases which are not clear represent mot-à-mot translations: the result in French does not make sense because of the fixed word order.
CML 2010 DUBROVNIK
EXPERIMENT: RESULTS AND ANALYSIS
Мишката Котанчо хвана и изчезна под дивана.The mouse Kotantcho caught and disappeared under the sofa
X(noun) VerbY(noun)
The mouse
Caught(activity)
Kotantcho (cat)
‘translate the story’ condition
SUBJECT Verb OBJECT
An attempt to comply with the original word order is seen in a couple of cases where the “subject first” rule is respected, leading to the translation of sentences in a passive form
“La sourris a été par Kotantcho attrapée” (The mouse has been by the cat caught).
CML 2010 DUBROVNIK
Following our model, the cognitive system first assembles a fractional representation of the sentence-meaning structure (coupled words for example) and uses working memory loops for checking the semantic consistency.
EXPERIMENT : ANALYSIS
Long
Term
Memory
ConceptsEvents
verbs
Attributes
Semantic primitives
Syntactic categories prep.
Semantic Operators
Syntactic rules
….
…
nouns adjectives
SentenceWorking memory
Z24
Z25
Z26
Z27
λ
l81
l105
l70l89
l87
l85 l86
l84
l83
l82
l100
l88
l94
l95
l96
l91
l92
l93
l90
λ
l12
l69
l49
φ φ
φ
CML 2010 DUBROVNIK
Z24
Z25
Z26
Z27
λ
l81
l105
l70l89
l87
l85 l86
l84
l83
l82
l100
l88
l94
l95
l96
l91
l92
l93
l90
l12
l69
l49
φ φ
φ
Verb
Y
X
λ
The grammatical information is not sufficient to build the syntactic structure.
The treatment can continue only by applying other mechanisms.
verbYX
Concept Concept Achievement
EXPERIMENT : ANALYSIS
CML 2010 DUBROVNIK
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
verbYX
Mouse Cat Caught
Stage I : Merge Subject-Verb
M (X*V*) = [X*, V] (?O)
X* = X* Acts (?O)
TheMouseCaught
CML 2010 DUBROVNIK
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
verbYX Stage I : Merge Subject-Verb
M (Y*V*) = [Y*, V] (?O)
Y* = Y* Acts (?O)
TheCatCaught
Mouse Cat Caught
CML 2010 DUBROVNIK
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
verbYX Stage I : Merge Subject-Verb
M (X*V*) = [X*, V] (?O)
X* = X* Acts (?O)
TheMouseCaught
Stage II Merge Acting Subject with Object
The Mouse Caught the Cat
Mouse Cat Caught
The Cat is the Object of TheMouseCaught
REJECTED
CML 2010 DUBROVNIK
SYNTACTIC PROCEDURE WITH SEMANTIC MERGE
verbYX Stage I : Merge Subject-Verb
M (Y*V*) = [Y*, V] (?O)
Y* = Y* Acts (?O)
TheCatCaught
Mouse Cat Caught
Stage II Merge Acting Subject with Object
The CatCaught the Mouse
The Mouse is the Object of TheCatCaught
ACCEPTED
CML 2010 DUBROVNIK
CONCLUSIONS
Experimental evidence is provided in support of the argument-based model. The semantic role of entities (nouns) is primary in syntax.
The analysis provided here is based on the assumption that the syntactic treatment includes a Merge operation between the images of the concepts. The obtained formal description of the stages of syntactic treatment explains the experimental results.
CML 2010 DUBROVNIK
The Mouse is the Object of TheCatCaught
THANK YOU!