shalmaneser 1.2.0.rc1 → 1.2.0.rc2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/README.md +26 -8
- data/doc/SB_README +57 -0
- data/doc/exp_files_description.txt +160 -0
- data/doc/fred.pdf +0 -0
- data/doc/index.md +120 -0
- data/doc/salsa_tool.pdf +0 -0
- data/doc/salsatigerxml.pdf +0 -0
- data/doc/shal_doc.pdf +0 -0
- data/doc/shal_lrec.pdf +0 -0
- data/lib/ext/maxent/Classify.class +0 -0
- data/lib/ext/maxent/Train.class +0 -0
- data/lib/frprep/TreetaggerInterface.rb +4 -4
- data/lib/shalmaneser/version.rb +1 -1
- metadata +41 -48
- data/test/frprep/test_opt_parser.rb +0 -94
- data/test/functional/functional_test_helper.rb +0 -40
- data/test/functional/sample_experiment_files/fred_test.salsa.erb +0 -122
- data/test/functional/sample_experiment_files/fred_train.salsa.erb +0 -135
- data/test/functional/sample_experiment_files/prp_test.salsa.erb +0 -138
- data/test/functional/sample_experiment_files/prp_test.salsa.fred.standalone.erb +0 -120
- data/test/functional/sample_experiment_files/prp_test.salsa.rosy.standalone.erb +0 -120
- data/test/functional/sample_experiment_files/prp_train.salsa.erb +0 -138
- data/test/functional/sample_experiment_files/prp_train.salsa.fred.standalone.erb +0 -138
- data/test/functional/sample_experiment_files/prp_train.salsa.rosy.standalone.erb +0 -138
- data/test/functional/sample_experiment_files/rosy_test.salsa.erb +0 -257
- data/test/functional/sample_experiment_files/rosy_train.salsa.erb +0 -259
- data/test/functional/test_fred.rb +0 -47
- data/test/functional/test_frprep.rb +0 -52
- data/test/functional/test_rosy.rb +0 -40
@@ -1,138 +0,0 @@
|
|
1
|
-
#################################################
|
2
|
-
# This is a sample experiment file
|
3
|
-
# with explanations of all features
|
4
|
-
# that can be set for the frprep preprocessing system for Fred and Rosy.
|
5
|
-
#
|
6
|
-
# To start your own experiment,
|
7
|
-
# replace all occurrences of
|
8
|
-
# %...% by values of your choice.
|
9
|
-
#
|
10
|
-
# Boolean features may be omitted and are false by default.
|
11
|
-
#
|
12
|
-
# Experiment file lines that start with '#'
|
13
|
-
# are comments and are ignored. Empty lines are ignored as well.
|
14
|
-
|
15
|
-
########################
|
16
|
-
# Experiment description
|
17
|
-
#
|
18
|
-
|
19
|
-
# ID identifying this experiment and all its data
|
20
|
-
# please do not use spaces inside the experiment ID
|
21
|
-
prep_experiment_ID = prp_train
|
22
|
-
|
23
|
-
# YOUR INPUT DATA:
|
24
|
-
# frprep accepts an input directory rather than an input file.
|
25
|
-
# It will process all files in the directory directory_input
|
26
|
-
# and write the results to directory_preprocessed.
|
27
|
-
#
|
28
|
-
# For input formats see the discussion of "format" below.
|
29
|
-
directory_input = <%= File.expand_path('test/functional/input/frprep/train.salsa') %>
|
30
|
-
directory_preprocessed = <%= File.expand_path('test/functional/output/frprep/train.salsa') %>
|
31
|
-
|
32
|
-
##
|
33
|
-
# Experimental data is described by the following parameters:
|
34
|
-
#
|
35
|
-
# - language: en / de
|
36
|
-
# en for English or de for German
|
37
|
-
#
|
38
|
-
# - format: SalsaTigerXML / FNXml / SalsaTab / BNC / Plain
|
39
|
-
#
|
40
|
-
# Format of the input data, training/test set
|
41
|
-
# SalsaTigerXML: Parsed data, English or German
|
42
|
-
# FNXml: FrameNet Lexical Unit files in FrameNet XML format
|
43
|
-
# FNCorpusXML: FrameNet files in the FrameNet corpus XML format
|
44
|
-
# SalsaTab: tabular format (internal)
|
45
|
-
# BNC BNC XML format, alternating words and POS tags
|
46
|
-
# Plain Plain text, ONE SENTENCE PER LINE.
|
47
|
-
#
|
48
|
-
# Preprocessing transforms all data to SalsaTigerXML.
|
49
|
-
#
|
50
|
-
# - origin: SalsaTiger / FrameNet / <not specified>
|
51
|
-
# This is the origin of the training/test data.
|
52
|
-
# SalsaTiger: data from the Tiger corpus, possibly semantically
|
53
|
-
# annotated by Salsa
|
54
|
-
# FrameNet: data from the FrameNet project
|
55
|
-
#
|
56
|
-
# Don't set 'origin' if none of these origins apply
|
57
|
-
#
|
58
|
-
# - encoding: utf8 / iso / hex / <not specified>
|
59
|
-
# Default: iso
|
60
|
-
|
61
|
-
language = de
|
62
|
-
#origin =
|
63
|
-
format = SalsaTigerXML
|
64
|
-
encoding = utf8
|
65
|
-
|
66
|
-
#############################
|
67
|
-
# Which preprocessing steps to take?
|
68
|
-
#
|
69
|
-
# Data can be parsed, lemmatized and POS-tagged,
|
70
|
-
# but this happens only if it is specified in the
|
71
|
-
# experiment file.
|
72
|
-
#
|
73
|
-
# Set these booleans to true to trigger the respective
|
74
|
-
# type of preprocessing. The default value is false.
|
75
|
-
|
76
|
-
do_lemmatize = true
|
77
|
-
do_postag = false
|
78
|
-
do_parse = true
|
79
|
-
|
80
|
-
#############################
|
81
|
-
# directory where frprep puts its internal data
|
82
|
-
#
|
83
|
-
|
84
|
-
frprep_directory = <%= File.expand_path('test/functional/output/') %>
|
85
|
-
|
86
|
-
#############################
|
87
|
-
# Syntax/semantics interface repair:
|
88
|
-
# FrameNet annotated data has some annotation choices
|
89
|
-
# that may make it harder to learn the mapping from
|
90
|
-
# syntactic structure to semantic roles.
|
91
|
-
#
|
92
|
-
# If you are using FrameNet data for training a
|
93
|
-
# semantic role labeler, set the following two settings
|
94
|
-
# to true (default is false) to 'repair' semantic role labels
|
95
|
-
# to closer match the syntactic structure
|
96
|
-
|
97
|
-
fe_syn_repair = true
|
98
|
-
fe_rel_repair = false
|
99
|
-
|
100
|
-
|
101
|
-
#################
|
102
|
-
# Location of tools and resources used by Fred
|
103
|
-
|
104
|
-
# currently known to the system:
|
105
|
-
# (Saarbruecken paths given)
|
106
|
-
#
|
107
|
-
# - POS tagging:
|
108
|
-
# - pos_tagger = treetagger
|
109
|
-
# pos_tagger_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
110
|
-
#
|
111
|
-
# - Lemmatization:
|
112
|
-
# - lemmatizer = treetagger
|
113
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
114
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-german-notokenisation
|
115
|
-
#
|
116
|
-
# - Parser:
|
117
|
-
# - parser = collins (English)
|
118
|
-
# parser_path = /proj/llx/Software/Parsers/COLLINS-PARSER/
|
119
|
-
# - parser = sleepy (German)
|
120
|
-
# parser_path = /proj/corpora/sleepy3/
|
121
|
-
# - parser = minipar (English)
|
122
|
-
# parser_path = /proj/llx/Software/Parsers/minipar-linux/
|
123
|
-
#
|
124
|
-
pos_tagger = treetagger
|
125
|
-
pos_tagger_path = <%= File.expand_path('tools/treetagger') %>
|
126
|
-
|
127
|
-
lemmatizer = treetagger
|
128
|
-
lemmatizer_path = <%= File.expand_path('tools/treetagger') %>
|
129
|
-
|
130
|
-
parser = berkeley
|
131
|
-
parser_path = <%= File.expand_path('tools/berkeleyParser') %>
|
132
|
-
|
133
|
-
# parser:
|
134
|
-
# maximum no. of sentences in a parse file,
|
135
|
-
# maximum sentence length to be parsed
|
136
|
-
|
137
|
-
parser_max_sent_num = 2000
|
138
|
-
parser_max_sent_len = 80
|
@@ -1,138 +0,0 @@
|
|
1
|
-
#################################################
|
2
|
-
# This is a sample experiment file
|
3
|
-
# with explanations of all features
|
4
|
-
# that can be set for the frprep preprocessing system for Fred and Rosy.
|
5
|
-
#
|
6
|
-
# To start your own experiment,
|
7
|
-
# replace all occurrences of
|
8
|
-
# %...% by values of your choice.
|
9
|
-
#
|
10
|
-
# Boolean features may be omitted and are false by default.
|
11
|
-
#
|
12
|
-
# Experiment file lines that start with '#'
|
13
|
-
# are comments and are ignored. Empty lines are ignored as well.
|
14
|
-
|
15
|
-
########################
|
16
|
-
# Experiment description
|
17
|
-
#
|
18
|
-
|
19
|
-
# ID identifying this experiment and all its data
|
20
|
-
# please do not use spaces inside the experiment ID
|
21
|
-
prep_experiment_ID = prp_train
|
22
|
-
|
23
|
-
# YOUR INPUT DATA:
|
24
|
-
# frprep accepts an input directory rather than an input file.
|
25
|
-
# It will process all files in the directory directory_input
|
26
|
-
# and write the results to directory_preprocessed.
|
27
|
-
#
|
28
|
-
# For input formats see the discussion of "format" below.
|
29
|
-
#directory_input = <%= File.expand_path('test/functional/input/frprep/train.salsa') %>
|
30
|
-
directory_preprocessed = <%= File.expand_path('test/functional/input/fred/frprep/train.salsa') %>
|
31
|
-
|
32
|
-
##
|
33
|
-
# Experimental data is described by the following parameters:
|
34
|
-
#
|
35
|
-
# - language: en / de
|
36
|
-
# en for English or de for German
|
37
|
-
#
|
38
|
-
# - format: SalsaTigerXML / FNXml / SalsaTab / BNC / Plain
|
39
|
-
#
|
40
|
-
# Format of the input data, training/test set
|
41
|
-
# SalsaTigerXML: Parsed data, English or German
|
42
|
-
# FNXml: FrameNet Lexical Unit files in FrameNet XML format
|
43
|
-
# FNCorpusXML: FrameNet files in the FrameNet corpus XML format
|
44
|
-
# SalsaTab: tabular format (internal)
|
45
|
-
# BNC BNC XML format, alternating words and POS tags
|
46
|
-
# Plain Plain text, ONE SENTENCE PER LINE.
|
47
|
-
#
|
48
|
-
# Preprocessing transforms all data to SalsaTigerXML.
|
49
|
-
#
|
50
|
-
# - origin: SalsaTiger / FrameNet / <not specified>
|
51
|
-
# This is the origin of the training/test data.
|
52
|
-
# SalsaTiger: data from the Tiger corpus, possibly semantically
|
53
|
-
# annotated by Salsa
|
54
|
-
# FrameNet: data from the FrameNet project
|
55
|
-
#
|
56
|
-
# Don't set 'origin' if none of these origins apply
|
57
|
-
#
|
58
|
-
# - encoding: utf8 / iso / hex / <not specified>
|
59
|
-
# Default: iso
|
60
|
-
|
61
|
-
language = de
|
62
|
-
#origin =
|
63
|
-
format = SalsaTigerXML
|
64
|
-
encoding = utf8
|
65
|
-
|
66
|
-
#############################
|
67
|
-
# Which preprocessing steps to take?
|
68
|
-
#
|
69
|
-
# Data can be parsed, lemmatized and POS-tagged,
|
70
|
-
# but this happens only if it is specified in the
|
71
|
-
# experiment file.
|
72
|
-
#
|
73
|
-
# Set these booleans to true to trigger the respective
|
74
|
-
# type of preprocessing. The default value is false.
|
75
|
-
|
76
|
-
do_lemmatize = true
|
77
|
-
do_postag = false
|
78
|
-
do_parse = true
|
79
|
-
|
80
|
-
#############################
|
81
|
-
# directory where frprep puts its internal data
|
82
|
-
#
|
83
|
-
|
84
|
-
#frprep_directory = <%= File.expand_path('test/functional/input/fred/') %>
|
85
|
-
|
86
|
-
#############################
|
87
|
-
# Syntax/semantics interface repair:
|
88
|
-
# FrameNet annotated data has some annotation choices
|
89
|
-
# that may make it harder to learn the mapping from
|
90
|
-
# syntactic structure to semantic roles.
|
91
|
-
#
|
92
|
-
# If you are using FrameNet data for training a
|
93
|
-
# semantic role labeler, set the following two settings
|
94
|
-
# to true (default is false) to 'repair' semantic role labels
|
95
|
-
# to closer match the syntactic structure
|
96
|
-
|
97
|
-
fe_syn_repair = true
|
98
|
-
fe_rel_repair = false
|
99
|
-
|
100
|
-
|
101
|
-
#################
|
102
|
-
# Location of tools and resources used by Fred
|
103
|
-
|
104
|
-
# currently known to the system:
|
105
|
-
# (Saarbruecken paths given)
|
106
|
-
#
|
107
|
-
# - POS tagging:
|
108
|
-
# - pos_tagger = treetagger
|
109
|
-
# pos_tagger_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
110
|
-
#
|
111
|
-
# - Lemmatization:
|
112
|
-
# - lemmatizer = treetagger
|
113
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
114
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-german-notokenisation
|
115
|
-
#
|
116
|
-
# - Parser:
|
117
|
-
# - parser = collins (English)
|
118
|
-
# parser_path = /proj/llx/Software/Parsers/COLLINS-PARSER/
|
119
|
-
# - parser = sleepy (German)
|
120
|
-
# parser_path = /proj/corpora/sleepy3/
|
121
|
-
# - parser = minipar (English)
|
122
|
-
# parser_path = /proj/llx/Software/Parsers/minipar-linux/
|
123
|
-
#
|
124
|
-
pos_tagger = treetagger
|
125
|
-
pos_tagger_path = <%= File.expand_path('tools/treetagger') %>
|
126
|
-
|
127
|
-
lemmatizer = treetagger
|
128
|
-
lemmatizer_path = <%= File.expand_path('tools/treetagger') %>
|
129
|
-
|
130
|
-
parser = berkeley
|
131
|
-
parser_path = <%= File.expand_path('tools/berkeleyParser') %>
|
132
|
-
|
133
|
-
# parser:
|
134
|
-
# maximum no. of sentences in a parse file,
|
135
|
-
# maximum sentence length to be parsed
|
136
|
-
|
137
|
-
parser_max_sent_num = 2000
|
138
|
-
parser_max_sent_len = 80
|
@@ -1,138 +0,0 @@
|
|
1
|
-
#################################################
|
2
|
-
# This is a sample experiment file
|
3
|
-
# with explanations of all features
|
4
|
-
# that can be set for the frprep preprocessing system for Fred and Rosy.
|
5
|
-
#
|
6
|
-
# To start your own experiment,
|
7
|
-
# replace all occurrences of
|
8
|
-
# %...% by values of your choice.
|
9
|
-
#
|
10
|
-
# Boolean features may be omitted and are false by default.
|
11
|
-
#
|
12
|
-
# Experiment file lines that start with '#'
|
13
|
-
# are comments and are ignored. Empty lines are ignored as well.
|
14
|
-
|
15
|
-
########################
|
16
|
-
# Experiment description
|
17
|
-
#
|
18
|
-
|
19
|
-
# ID identifying this experiment and all its data
|
20
|
-
# please do not use spaces inside the experiment ID
|
21
|
-
prep_experiment_ID = prp_train
|
22
|
-
|
23
|
-
# YOUR INPUT DATA:
|
24
|
-
# frprep accepts an input directory rather than an input file.
|
25
|
-
# It will process all files in the directory directory_input
|
26
|
-
# and write the results to directory_preprocessed.
|
27
|
-
#
|
28
|
-
# For input formats see the discussion of "format" below.
|
29
|
-
#directory_input = <%= File.expand_path('test/functional/input/frprep/train.salsa') %>
|
30
|
-
directory_preprocessed = <%= File.expand_path('test/functional/input/rosy/frprep/train.salsa') %>
|
31
|
-
|
32
|
-
##
|
33
|
-
# Experimental data is described by the following parameters:
|
34
|
-
#
|
35
|
-
# - language: en / de
|
36
|
-
# en for English or de for German
|
37
|
-
#
|
38
|
-
# - format: SalsaTigerXML / FNXml / SalsaTab / BNC / Plain
|
39
|
-
#
|
40
|
-
# Format of the input data, training/test set
|
41
|
-
# SalsaTigerXML: Parsed data, English or German
|
42
|
-
# FNXml: FrameNet Lexical Unit files in FrameNet XML format
|
43
|
-
# FNCorpusXML: FrameNet files in the FrameNet corpus XML format
|
44
|
-
# SalsaTab: tabular format (internal)
|
45
|
-
# BNC BNC XML format, alternating words and POS tags
|
46
|
-
# Plain Plain text, ONE SENTENCE PER LINE.
|
47
|
-
#
|
48
|
-
# Preprocessing transforms all data to SalsaTigerXML.
|
49
|
-
#
|
50
|
-
# - origin: SalsaTiger / FrameNet / <not specified>
|
51
|
-
# This is the origin of the training/test data.
|
52
|
-
# SalsaTiger: data from the Tiger corpus, possibly semantically
|
53
|
-
# annotated by Salsa
|
54
|
-
# FrameNet: data from the FrameNet project
|
55
|
-
#
|
56
|
-
# Don't set 'origin' if none of these origins apply
|
57
|
-
#
|
58
|
-
# - encoding: utf8 / iso / hex / <not specified>
|
59
|
-
# Default: iso
|
60
|
-
|
61
|
-
language = de
|
62
|
-
#origin =
|
63
|
-
format = SalsaTigerXML
|
64
|
-
encoding = utf8
|
65
|
-
|
66
|
-
#############################
|
67
|
-
# Which preprocessing steps to take?
|
68
|
-
#
|
69
|
-
# Data can be parsed, lemmatized and POS-tagged,
|
70
|
-
# but this happens only if it is specified in the
|
71
|
-
# experiment file.
|
72
|
-
#
|
73
|
-
# Set these booleans to true to trigger the respective
|
74
|
-
# type of preprocessing. The default value is false.
|
75
|
-
|
76
|
-
do_lemmatize = true
|
77
|
-
do_postag = false
|
78
|
-
do_parse = true
|
79
|
-
|
80
|
-
#############################
|
81
|
-
# directory where frprep puts its internal data
|
82
|
-
#
|
83
|
-
|
84
|
-
frprep_directory = <%= File.expand_path('test/functional/input/rosy/') %>
|
85
|
-
|
86
|
-
#############################
|
87
|
-
# Syntax/semantics interface repair:
|
88
|
-
# FrameNet annotated data has some annotation choices
|
89
|
-
# that may make it harder to learn the mapping from
|
90
|
-
# syntactic structure to semantic roles.
|
91
|
-
#
|
92
|
-
# If you are using FrameNet data for training a
|
93
|
-
# semantic role labeler, set the following two settings
|
94
|
-
# to true (default is false) to 'repair' semantic role labels
|
95
|
-
# to closer match the syntactic structure
|
96
|
-
|
97
|
-
fe_syn_repair = true
|
98
|
-
fe_rel_repair = false
|
99
|
-
|
100
|
-
|
101
|
-
#################
|
102
|
-
# Location of tools and resources used by Fred
|
103
|
-
|
104
|
-
# currently known to the system:
|
105
|
-
# (Saarbruecken paths given)
|
106
|
-
#
|
107
|
-
# - POS tagging:
|
108
|
-
# - pos_tagger = treetagger
|
109
|
-
# pos_tagger_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
110
|
-
#
|
111
|
-
# - Lemmatization:
|
112
|
-
# - lemmatizer = treetagger
|
113
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
|
114
|
-
# lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-german-notokenisation
|
115
|
-
#
|
116
|
-
# - Parser:
|
117
|
-
# - parser = collins (English)
|
118
|
-
# parser_path = /proj/llx/Software/Parsers/COLLINS-PARSER/
|
119
|
-
# - parser = sleepy (German)
|
120
|
-
# parser_path = /proj/corpora/sleepy3/
|
121
|
-
# - parser = minipar (English)
|
122
|
-
# parser_path = /proj/llx/Software/Parsers/minipar-linux/
|
123
|
-
#
|
124
|
-
pos_tagger = treetagger
|
125
|
-
pos_tagger_path = <%= File.expand_path('tools/treetagger') %>
|
126
|
-
|
127
|
-
lemmatizer = treetagger
|
128
|
-
lemmatizer_path = <%= File.expand_path('tools/treetagger') %>
|
129
|
-
|
130
|
-
parser = berkeley
|
131
|
-
parser_path = <%= File.expand_path('tools/berkeleyParser') %>
|
132
|
-
|
133
|
-
# parser:
|
134
|
-
# maximum no. of sentences in a parse file,
|
135
|
-
# maximum sentence length to be parsed
|
136
|
-
|
137
|
-
parser_max_sent_num = 2000
|
138
|
-
parser_max_sent_len = 80
|
@@ -1,257 +0,0 @@
|
|
1
|
-
#################################################
|
2
|
-
# This is a sample experiment file
|
3
|
-
# with explanations of all features
|
4
|
-
# that can be set for the ROSY system.
|
5
|
-
#
|
6
|
-
# To start your own experiment,
|
7
|
-
# replace all occurrences of
|
8
|
-
# %SOMETHING% or %PATH% or %PARAMETERS%
|
9
|
-
# by values of your choice.
|
10
|
-
#
|
11
|
-
# Experiment file lines that start with '#'
|
12
|
-
# are comments and are ignored. Empty lines are ignored as well.
|
13
|
-
|
14
|
-
########################
|
15
|
-
# Experiment description
|
16
|
-
#
|
17
|
-
|
18
|
-
##
|
19
|
-
# Experiment ID:
|
20
|
-
# Uniquely identifies files and database tables
|
21
|
-
# of this experiment.
|
22
|
-
# The experiment ID is a word (no spaces) of
|
23
|
-
# letters in [A-Za-z_].
|
24
|
-
experiment_ID = rosy_test
|
25
|
-
|
26
|
-
# Enduser mode?
|
27
|
-
# The idea is that the enduser will only _apply_
|
28
|
-
# pre-trained classifiers. So in enduser mode many
|
29
|
-
# options are disallowed.
|
30
|
-
enduser_mode = false
|
31
|
-
|
32
|
-
# directories
|
33
|
-
# - data directory: where Rosy puts its internal data
|
34
|
-
# - input directory:
|
35
|
-
# where Rosy reads its input SalsaTigerXML data.
|
36
|
-
# One directory each for the training and the test data
|
37
|
-
# - output directory:
|
38
|
-
# where Rosy writes its output SalsaTigerXML data:
|
39
|
-
# same frames as in the input data, but frame elements newly
|
40
|
-
# assigned.
|
41
|
-
# If no output directory is given, output is to
|
42
|
-
# <data_dir>/<experiment_ID>/output/
|
43
|
-
# - classifier_dir: If present, this is where trained classifiers
|
44
|
-
# are written.
|
45
|
-
# Otherwise they are written to <data_dir>/<experiment_id>/classif_dir
|
46
|
-
data_dir = <%= File.expand_path('test/functional/output') %>
|
47
|
-
directory_input_test = <%= File.expand_path('test/functional/input/rosy/test.salsa') %>
|
48
|
-
classifier_dir = <%= File.expand_path('test/functional/input/rosy/cls') %>
|
49
|
-
|
50
|
-
##
|
51
|
-
# Preprocessing settings:
|
52
|
-
# frprep experiment files for training and test data.
|
53
|
-
preproc_descr_file_train = <%= File.expand_path('test/functional/sample_experiment_files/prp_train.salsa.rosy.standalone') %>
|
54
|
-
preproc_descr_file_test = <%= File.expand_path('test/functional/sample_experiment_files/prp_test.salsa.rosy.standalone') %>
|
55
|
-
|
56
|
-
|
57
|
-
########################
|
58
|
-
# features
|
59
|
-
#
|
60
|
-
# Please specify all features that you would like
|
61
|
-
# Rosy to compute.
|
62
|
-
# Note: The system distinguishes between features to be
|
63
|
-
# computed and features to be included in the model,
|
64
|
-
# so you can compute features once and then vary features
|
65
|
-
# included in the model.
|
66
|
-
#
|
67
|
-
# Format for each feature specification:
|
68
|
-
# feature = <feature_name> [dontuse | argrec | arglab | onestep]
|
69
|
-
#
|
70
|
-
# dontuse: the feature is computed but not included in the model.
|
71
|
-
# argrec, arglab, onestep: the feature is used only in this
|
72
|
-
# processing step
|
73
|
-
#
|
74
|
-
#
|
75
|
-
# The set of features computed must stay the same throughout
|
76
|
-
# an experiment (or the match of experiment file and
|
77
|
-
# database table will fail), but the set of features included
|
78
|
-
# in the model can be varied.
|
79
|
-
#
|
80
|
-
# See below for a list of all features currently available in the system.
|
81
|
-
|
82
|
-
feature = pt_path
|
83
|
-
feature = gf_path
|
84
|
-
feature = path
|
85
|
-
feature = path_length
|
86
|
-
feature = pt_combined_path
|
87
|
-
feature = gf_combined_path
|
88
|
-
feature = combined_path
|
89
|
-
feature = pt_partial_path
|
90
|
-
feature = gf_partial_path
|
91
|
-
feature = partial_path
|
92
|
-
feature = pt_gvpath
|
93
|
-
feature = gf_gvpath
|
94
|
-
feature = gvpath
|
95
|
-
feature = ancestor_rule
|
96
|
-
feature = relpos
|
97
|
-
feature = pt
|
98
|
-
feature = gf
|
99
|
-
feature = father_pt
|
100
|
-
feature = frame
|
101
|
-
feature = target
|
102
|
-
feature = target_pos
|
103
|
-
feature = target_voice
|
104
|
-
feature = gov_verb
|
105
|
-
feature = prep
|
106
|
-
feature = const_head
|
107
|
-
feature = const_head_pos
|
108
|
-
feature = icont_word
|
109
|
-
feature = firstword
|
110
|
-
feature = lastword
|
111
|
-
feature = leftsib
|
112
|
-
feature = rightsib
|
113
|
-
feature = worddistance
|
114
|
-
feature = ismaxproj
|
115
|
-
feature = nearest_node
|
116
|
-
feature = prune
|
117
|
-
|
118
|
-
########################
|
119
|
-
# classifiers
|
120
|
-
#
|
121
|
-
# Please specify each classifier type you want to use.
|
122
|
-
# If you specify more than one classifier, classifier combination
|
123
|
-
# is used.
|
124
|
-
#
|
125
|
-
# Format for each classifier specification:
|
126
|
-
# classifier = <classifier_name> <path> [<parameters>]
|
127
|
-
#
|
128
|
-
# Possible values for <classifier_name> at the moment:
|
129
|
-
# timbl (memory-based learning),
|
130
|
-
# maxent (openlp maxent system)
|
131
|
-
#
|
132
|
-
# Samples:
|
133
|
-
# classifier = timbl /prog/MachineLearning/Timbl5/
|
134
|
-
# classifier = maxent /prog/maxent-2.4.0 /prog/shalmaneser/program/tools/maxent
|
135
|
-
|
136
|
-
classifier = maxent <%= File.expand_path('tools/maxent/maxent-2.4.0') %>
|
137
|
-
|
138
|
-
########################
|
139
|
-
# further settings
|
140
|
-
|
141
|
-
# Pruning: Identify constituents that are very unlikely
|
142
|
-
# to instantiate a semantic role, and prune them prior
|
143
|
-
# to the training/application of classifiers?
|
144
|
-
#
|
145
|
-
# Pruning methods available at the moment:
|
146
|
-
# prune: Xue/Palmer EMNLP 2004, adapted to fit each individual parser
|
147
|
-
#
|
148
|
-
# To enable pruning, set "prune" to the pruning method of your choice,
|
149
|
-
# and also compute the feature of the same name -- see
|
150
|
-
# feature list below.
|
151
|
-
# To disable pruning, comment out the next line.
|
152
|
-
prune = prune
|
153
|
-
|
154
|
-
# verbose mode
|
155
|
-
verbose = true
|
156
|
-
|
157
|
-
# data adaptation:
|
158
|
-
# correct training labels to
|
159
|
-
# match syntax better?
|
160
|
-
fe_syn_repair = true
|
161
|
-
fe_rel_repair = false
|
162
|
-
|
163
|
-
# xwise: For each classification step (argrec, arglab, onestep)
|
164
|
-
# you can set the granularity of training:
|
165
|
-
# - by frame (frame)
|
166
|
-
# - by target part of speech or (target_pos)
|
167
|
-
# - by target lemma. (target)
|
168
|
-
#
|
169
|
-
# these three settings can be combined, e.g.
|
170
|
-
# xwise_argrec = target_pos frame
|
171
|
-
# to train argrec frame-wise and split each frame by target POS.
|
172
|
-
#
|
173
|
-
# If no value is given for xwise_<step>, the default is "frame".
|
174
|
-
xwise_argrec = frame
|
175
|
-
xwise_arglab = frame
|
176
|
-
xwise_onestep = frame
|
177
|
-
|
178
|
-
|
179
|
-
# assume_argrec_perfect: by default, this is false.
|
180
|
-
#
|
181
|
-
# Set this to true
|
182
|
-
# to perform the arglab (argument labeling) step
|
183
|
-
# on all instances that actually are FEs
|
184
|
-
# rather than on all instances that the argrec step
|
185
|
-
# has judged to be FEs.
|
186
|
-
assume_argrec_perfect = false
|
187
|
-
|
188
|
-
# split_nones: set to true
|
189
|
-
# to split the NONE target class into:
|
190
|
-
# NONE left of target,
|
191
|
-
# NONE right of target
|
192
|
-
# because the NONE class has so many more instances
|
193
|
-
# than any other.
|
194
|
-
split_nones = true
|
195
|
-
|
196
|
-
|
197
|
-
# print_eval_log: set to true to print individual correctness
|
198
|
-
# judgments for each instance evaluated
|
199
|
-
print_eval_log = true
|
200
|
-
|
201
|
-
# External data source:
|
202
|
-
#
|
203
|
-
# Rosy can integrate data computed by additional systems
|
204
|
-
# provided that they all use a common experiment file
|
205
|
-
# for external data to determine where they put their data.
|
206
|
-
# Rosy needs the path to that experiment file.
|
207
|
-
#
|
208
|
-
# (May be left unset when no external data is used)
|
209
|
-
#external_descr_file = %PATH%
|
210
|
-
|
211
|
-
|
212
|
-
########################
|
213
|
-
# rosy internal data - please don't change
|
214
|
-
|
215
|
-
# Database access:
|
216
|
-
# dbtype: type of database, either mysql
|
217
|
-
# for a MySQL server, or sqlite for SQLite.
|
218
|
-
#
|
219
|
-
# if dbtype == mysql, set access parameters:
|
220
|
-
# host: database server
|
221
|
-
# user: user name to use
|
222
|
-
# passwd: password for user
|
223
|
-
# dbname: database where all Rosy's tables will be stored
|
224
|
-
|
225
|
-
dbtype = mysql
|
226
|
-
host = localhost
|
227
|
-
user = shalm
|
228
|
-
passwd = 12345
|
229
|
-
dbname = shalm11
|
230
|
-
|
231
|
-
# classifier output columns in the tables all start
|
232
|
-
# with this prefix
|
233
|
-
classif_column_name = classif
|
234
|
-
|
235
|
-
# pattern for constructing the names
|
236
|
-
# of the DB tables with training data (main_table_name)
|
237
|
-
# and test data (test_table_name)
|
238
|
-
main_table_name = rosy_<exp_ID>_main
|
239
|
-
test_table_name = rosy_<exp_ID>_<test_ID>
|
240
|
-
|
241
|
-
# string to use for "no value for this feature"
|
242
|
-
# as well as "no FE for this instance"
|
243
|
-
noval = NONE
|
244
|
-
|
245
|
-
# pattern for constructing the names
|
246
|
-
# of classifier files and classifier output files
|
247
|
-
classifier_file = classif.<classif>.<group>
|
248
|
-
classifier_output_file = classout.<classif>.<group>.<dataset>
|
249
|
-
|
250
|
-
# pattern for constructing the names
|
251
|
-
# of the evaluation file and the evaluation log file
|
252
|
-
eval_file = eval.<exp_ID>.<step>.<test_ID>
|
253
|
-
log_file = eval_log.<exp_ID>.<step>.<test_ID>
|
254
|
-
|
255
|
-
# pattern for constructing the names
|
256
|
-
# of the files with failed parses
|
257
|
-
failed_file = parsefail.<exp_ID>.<split_ID>.<dataset>
|