Word list entry: (spiser (V spise Pres)) Stem list entry: (spise (V Transitive (sense eat'))) Template list entries: (V ((sense) (trans relation))) (Pres ((syntax form) finite) ((syntax tense) present) ((trans loc cond relation) temp-overlap) ((trans loc cond arg1 ind) LOCATION) ((trans loc cond arg2 ind) DISCOURSE-LOC) ((trans loc ind) (trans loc cond arg1 ind))) (Transitive ((syntax transitive) yes) ((syntax subj trans) (trans arg1)) ((syntax obj trans) (trans arg2)))
Word list entry: (spiser (V spise Pres)) Stem list entry: (spise (V Transitive (sense eat'))) Template list entries: (V ((sense) (trans relation))) (Pres((syntax.
Mar 31, 2015
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Word list entry:(spiser (V spise Pres))
Stem list entry:(spise (V Transitive (sense eat')))
Template list entries:(V ((sense) (trans relation)))(Pres ((syntax form) finite)
((syntax tense) present)((trans loc cond relation) temp-overlap)((trans loc cond arg1 ind) LOCATION)((trans loc cond arg2 ind) DISCOURSE-LOC)((trans loc ind) (trans loc cond arg1 ind)))
(Transitive ((syntax transitive) yes)((syntax subj trans) (trans arg1))((syntax obj trans) (trans arg2)))
Word list entry:(spiser (V spise Pres))
Stem list entry:(spise (V Transitive (sense eat')))
Template list entries:(V ((sense) (trans relation)))(Pres ((syntax form) finite)
((syntax tense) present)((trans loc cond relation) temp-overlap)((trans loc cond arg1 ind) LOCATION)((trans loc cond arg2 ind) DISCOURSE-LOC)((trans loc ind) (trans loc cond arg1 ind)))
(Transitive ((syntax transitive) yes)((syntax subj trans) (trans arg1))((syntax obj trans) (trans arg2)))
sense
form
tense
transitive
finite
present
yes
transsubj
obj trans
syntax
relation eat'
arg1 []
arg2 []
ind
relation temp-overlap
arg1 ind LOCATION
arg2 ind DISCOURSE-LOC
loc
trans
cond
(S NP VP (0 (2)) ((0 syntax subj) (1)))
(S NP VP (0 (2)) ((0 syntax subj) (1)))
S
NP
VP syntax subj []
(S NP VP (0 (2)) ((0 syntax subj) (1)))
S
NP
VP syntax subj []
Situasjonsskjema for ”A man sees Mary”:
(S NP VP (0 (2)) ((0 syntax subj) (1)))
S
NP
VP syntax subj []
relation see
ind
cond
arg1 relation man
arg1
pol 1
ind1
arg2 Mary
loc
ind
cond
relation temp-overlap
arg1 ind2
arg2 ld
pol 1
Situasjonsskjema for ”A man sees Mary”:
Situasjonsskjema for ”A small ugly man sees Mary”:
relation
ind
cond
arg1
relation
arg1
arg2
loc
ind
cond
relation temp-overlap
arg1
arg2
small' yes
ugly' yes
yesman'
see' yes
named-Mary
ind
cond
yes
yes
INDIVIDUAL
INDIVIDUAL
LOCATION
DISC-LOC
arg1
relation
Situasjonsskjema for ”A small ugly man sees Mary”:
relation
arg1
arg2
LOCATION
yes
SPEAKER
arg2 HEARER
INFORM
relation leave' yes
ind
relation named-John yes
arg1 INDIVIDUAL
cond
arg1
loc
ind
relation temp-precede yes
arg1cond
arg3
loc DISC-LOC
ind
ind
Situasjonsskjema med ytringsinformasjon (”John left”):
syntax
trans
...
...
tense pres
subj
syntax
trans
syntax
obj
trans
...
Grunnstruktur i trrekkstruklturene:
relation
ind
cond
arg1 relation
arg1
arg2
loc
ind
cond
relation
arg1
arg2
see' yes
ind
yes
LOCATION
ind
named-John yes
named-Mary yesrelation
ind INDIVIDUALarg1
arg3
arg2
arg1
relation INFORM yes
SPEAKER
HEARER
loc DISC-LOC
trans
trans
tense pres
subj
obj
syntax
trans
syntax
trans
num sg
num sgsyntax
INDIVIDUAL
cond
temp-overlap
prop
syntax
#1
#1
Full struktur for”John sees Mary”:
Bottom-up parsing: only a partial analysis may result:
S' Srel VP
NP
PP
1 2 3 4 5 6 7 8 9 10 11 12 13
That system is preferable which translates sentences one chunk at a time
NP VP VP NP
Bottom-up parsing: only a partial analysis may result:
S' Srel VP
NP
PP
1 2 3 4 5 6 7 8 9 10 11 12 13
That system is preferable which translates sentences one chunk at a time
NP VP VP NP
Bottom-up parsing: only a partial analysis may result:
Maximal analyses are found:
S' Srel VP
NP
PP
1 2 3 4 5 6 7 8 9 10 11 12 13
That system is preferable which translates sentences one chunk at a time
NP VP VP NP
Bottom-up parsing: only a partial analysis may result:
S' Srel
NP
PP
1 5 8 10 13
That system is preferable which translates sentences one chunk at a time
Maximal analyses are found:
S' Srel VP
NP
PP
1 2 3 4 5 6 7 8 9 10 11 12 13
That system is preferable which translates sentences one chunk at a time
NP VP VP NP
Bottom-up parsing: only a partial analysis may result:
S' Srel
NP
PP
1 5 8 10 13
That system is preferable which translates sentences one chunk at a time
Maximal analyses are found:
Each edge (= tree) in each maximal analysis is translated separately
relation
ind
cond
arg1 relation
arg1
arg2
loc
ind
cond
relation
arg1
arg2
see' yes
ind
yes
LOCATION
ind
named-John yes
named-Mary yesrelation
ind INDIVIDUALarg1
arg3
arg2
arg1
relation INFORM yes
SPEAKER
HEARER
loc DISC-LOC
trans
trans
tense pres
subj
obj
syntax
trans
syntax
trans
num sg
num sgsyntax
INDIVIDUAL
cond
temp-overlap
prop
syntax
#1
#1
S'
S
NP
PropN
VP
V NP
PropNJohn sees
Mary
Result of a parse:tree with associated features:
ENGLISH NORWEGIAN
frighten skremme
Common cat
Common sense!
Common
arg
linkings
syntax
subj [trans ]
obj [trans ]
trans
arg1 []
arg2 []
ptrs [ nor [ mode1 69 ]]
sense frighten'
index 78
syntax
subj [trans ]
obj [trans ]
trans
arg1 []
arg2 []
sense frighten'
index 69
Resulting
added
information
cat V cat V
please like
syntax
subj [trans ]
obj [trans ]
trans
arg1 []
arg2 []
sense like'
index 156
cat V
syntax
subj [trans ]
obj [trans ]
trans
arg1 []
arg2 []
sense like'
index 149
cat V
Common cat
Common sense!
Different
arg
linkings
Result:
No pointer
added
LEXICONCOMPARISON
NP [ syntax [ poss
ENGLISH NORWEGIAN
S -> NP VP S -> NP VP
ptrs [ nor [ mode1 yes ]]
Resulting
added
information
(no pointer)
S [ syntax [ subj
NP
VP
index 3
Same mother
and dtr cats
Same dtr
sequence
Dtr structs
values of
same mthr
struct paths
index 5
NP -> POSS N' NP -> N' POSS
Resulting
added
information
index 12
Same mother
and dtr cats
Diff dtr
sequence
Dtr structs
values of
same mthr
struct paths
index 15
mode2 15
mode1 no
norptrs
[] ]]S [ syntax [ subj
NP
VP
[] ]]
NP [ syntax [ poss[]
]]
POSS
N'
[]]]
POSS
N'
GRAMMARCOMPARISON
TARGET LANGUAGESOURCE LANGUAGE
SITUATION
SCHEMA
SET OF
FULL
FEATURE
STRUCTURES
TOP DOWN
PREDICTION
OF RULES
AND STEMS
SEARCH FOR
COMPATIBLE
WORD FORMS
S O U R C E S T R I N G
TARGET
STRINGS
TREES AND
FEATURE STRUCTURES
WITH TARGET POINTERS
MODE CHECK3
SPLICE IN
TARGET SUBTREES
AT 2-NODES;
OVERWRITE FEATS
2
1INSERT TARGET
STEMS AT LEAF
NODES; OVER-
WRITE FEATS
BOTTOM-UP PARSING
For each tree
EXTRACT SITU-
ATION SCHEMA
FROM TOP NODE'S
FEATURE
STRUCTURE
SEARCH FOR TARGET
STEMS EXPRESSING
RELATIONS IN
SITUATION SCHEMA
SET OF SENSE-
CARRYING TARG
STEM-ENTRIES
UNIFY TARG
ENTRIES WITH
SITUATION
SCHEMA
TREES WITH
STEM ENTRIES
AT LEAF
NODES
The fulltranslation process
S
NP VP
PROPN V NP
DET N
Mozart frightens that dog
1
1
1 1
1
1
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
frighten'
[ "that dog" ]
[ "Mozart" ]
SOURCE LANGUAGE
TARGET LANGUAGE
Mozart PROPN [features]
den DET [features]
hund N [features]
skremme V [features]
Parse with target pointersMode 1
Output:"Mozart skremmer den hunden"
S
NP VP
PROPN V NP
DET N
Mozart skremme hundden
1
1
1 1
1
1
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
frighten'
[ "that dog" ]
[ "Mozart" ]
SOURCE LANGUAGE
TARGET LANGUAGE
Mozart PROPN [features]
den DET [features]
hund N [features]
skremme V [features]
Overwrite
Mode 1 and 2 generationOverwrite target stems at leaf nodes
S
NP VP
PROPN V NP
DET N
Mozart skremme hundden
1
1
1 1
1
1
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
frighten'
[ "that dog" ]
[ "Mozart" ]
skremme
skremmer
skremte
skremt
Mozart
Mozarts
den
det
de
dem
hund
hunden
hunder
hundene
hunds
hundens
hunders
hundenes
Unify strings
of forms with
terminal nodes
undoably; save
successful strings
Mode 1, 2 and 3 generationInsert compatible word forms at terminal nodes
Handled in Mode 1:
•Agreement•Source/target gender clashes•Syncretisms in source paradigms
S
NP VP
PROPN V NP
POSS N
Mozart frightens my dog
1
1
1 1
1
2
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
frighten'
[ "my dog" ]
[ "Mozart" ]
SOURCE LANGUAGE
TARGET LANGUAGE
NP -> N POSS
Mozart PROPN [features]
min POSS [features]
hund N [features]
skremme V [features]
Parse with target pointersMode 2
S
NP VP
PROPN V NP
POSSN
Mozart frightens mydog
1
1
1 1
1
2
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
frighten'
[ "my dog" ]
[ "Mozart" ]
SOURCE LANGUAGE
TARGET LANGUAGE
NP -> N POSS
Mozart PROPN [features]
min POSS [features]
hund N [features]
Splice
skremme V [features]
& overwrite
Mode 2 generationSplice in target subtrees
S
NP VP
PROPN V NP
POSS N
Mozart pleases my dog
1
1
1 3
1
2
1 1
tense present
syntax subj
obj
syntax
trans
[...]
syntax
trans
[...]
trans
relation
arg1
arg2
like'
[ "my dog" ]
[ "Mozart" ]
SOURCE LANGUAGE
TARGET LANGUAGE
NP -> N POSS
Mozart PROPN [features]
min POSS [features]
hund N [features]
Parse with target pointersMode 3
ind
relation temp-precede yes
arg1 ind LOCATION
arg2 ind DISC-LOC
cond
arg1
loc
relation request-truth-value yes
arg1 SPEAKER
arg2 HEARER
arg3
loc DISC-LOC
arg2
relation yeslike'
ind
relation yes
arg1 ind
cond
ind
relation yes
arg1 ind
cond
INDIVIDUAL
named-John
solution'
INDIVIDUAL
ref unique
ref unique
Mode 3 generation, Stage IInput situation schema:
form finite
mode interrogative
prop
syntax
relation request-truth-value yes
arg1 SPEAKER
arg2 HEARER
arg3
loc DISC-LOC
trans
S'
syntax
trans
#1
#1
[]
polarity
subj trans
obj trans
relation like' []
arg1
arg2
syntax
trans []
[]
fset
cat V
subcatrule
V'
V
NP
obj
[]
stem like
Target rule structure:
Target stem entry:
form finite
mode interrogative
prop
syntax
syntax
trans
#1
#1
ind
relation temp-precede yes
arg1 ind LOCATION
arg2 ind DISC-LOC
cond
arg1
loc
relation request-truth-value yes
arg1 SPEAKER
arg2 HEARER
arg3
loc DISC-LOC
arg2
relation yeslike'
ind
relation yes
arg1 ind
cond
ind
relation yes
arg1 ind
cond
INDIVIDUAL
named-John
solution'
INDIVIDUAL
ref unique
ref unique
trans
polarity
subj trans
obj trans
Result of unificationUnification pointsfor further structuresare identified
Set of initial feature structures:
S -> NP VP:
NP -> DET N':
?S
?NP ?VP
?DET ?N'
the
Top-down
prediction
of rules
constrained
by feature
structures
Mode 3 generation, Stage II
A
B C
D E
b1
b2
b3
d1
d2
e1
e2
e3
e4
Mode 3 generation, Stage IIIAlternative word forms at terminal nodes
Related Documents
