Datasets:

Languages:
English
ArXiv:
License:
system HF staff commited on
Commit
56b9abd
1 Parent(s): dd486cd

Update files from the datasets library (from 1.3.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.3.0

Files changed (4) hide show
  1. README.md +116 -142
  2. dataset_infos.json +1 -1
  3. dummy/0.0.0/dummy_data.zip +2 -2
  4. swda.py +677 -77
README.md CHANGED
@@ -43,11 +43,12 @@ task_ids:
43
  - [Dataset Curators](#dataset-curators)
44
  - [Licensing Information](#licensing-information)
45
  - [Citation Information](#citation-information)
 
46
 
47
  ## Dataset Description
48
 
49
  - **Homepage: [The Switchboard Dialog Act Corpus](http://compprag.christopherpotts.net/swda.html)**
50
- - **Repository: [NathanDuran/Switchboard-Corpus](https://github.com/NathanDuran/Switchboard-Corpus)**
51
  - **Paper:[The Switchboard Dialog Act Corpus](http://compprag.christopherpotts.net/swda.html)**
52
  = **Leaderboard: [Dialogue act classification](https://github.com/sebastianruder/NLP-progress/blob/master/english/dialogue.md#dialogue-act-classification)**
53
  - **Point of Contact: [Christopher Potts](https://web.stanford.edu/~cgpotts/)**
@@ -87,118 +88,100 @@ Utterance are tagged with the [SWBD-DAMSL](https://web.stanford.edu/~jurafsky/ws
87
 
88
  An example from the dataset is:
89
 
90
- `{'dialogue_act_tag': 17, 'speaker': 0, 'utterance_text': 'Okay.'}`
91
-
92
- where 17 correspond to `fo_o_fw_"_by_bc` (Other)
93
 
94
  ### Data Fields
95
- `speaker` - Refers to the current speaker talking. It is used to detect when a speaker change occurs.
96
- There are two values for speaker: `A` and `B`. This does not mean we only have tow speakers in the whole datasets.
97
- It's only used to signal if next utterance is from same speaker or from next speaker. Since we encoded all labels
98
- `A=0` and `B=1`.
99
-
100
- `utterance_text` - Text that a speaker says.
101
-
102
- `dialogue_act_tag` - Dialogue act label associated with the `utterance_text`. There are 41 dialogue act labels for this
103
- dataset. Each dialogue act label has a specific meaning:
104
-
105
- | Int | Dialogue Act | Labels |
106
- |-- |------------------------------ |----------------- |
107
- | 0 | Statement-non-opinion | sd |
108
- | 1 | Acknowledge (Backchannel) | b |
109
- | 2 | Statement-opinion | sv |
110
- | 3 | Uninterpretable | % |
111
- | 4 | Agree/Accept | aa |
112
- | 5 | Appreciation | ba |
113
- | 6 | Yes-No-Question | qy |
114
- | 7 | Yes Answers | ny |
115
- | 8 | Conventional-closing | fc |
116
- | 9 | Wh-Question | qw |
117
- | 10 | No Answers | nn |
118
- | 11 | Response Acknowledgement | bk |
119
- | 12 | Hedge | h |
120
- | 13 | Declarative Yes-No-Question | qy^d |
121
- | 14 | Backchannel in Question Form | bh |
122
- | 15 | Quotation | ^q |
123
- | 16 | Summarize/Reformulate | bf |
124
- | 17 | Other | fo_o_fw_"_by_bc |
125
- | 18 | Affirmative Non-yes Answers | na |
126
- | 19 | Action-directive | ad |
127
- | 20 | Collaborative Completion | ^2 |
128
- | 21 | Repeat-phrase | b^m |
129
- | 22 | Open-Question | qo |
130
- | 23 | Rhetorical-Question | qh |
131
- | 24 | Hold Before Answer/Agreement | ^h |
132
- | 25 | Reject | ar |
133
- | 26 | Negative Non-no Answers | ng |
134
- | 27 | Signal-non-understanding | br |
135
- | 28 | Other Answers | no |
136
- | 29 | Conventional-opening | fp |
137
- | 30 | Or-Clause | qrr |
138
- | 31 | Dispreferred Answers | arp_nd |
139
- | 32 | 3rd-party-talk | t3 |
140
- | 33 | Offers, Options Commits | oo_co_cc |
141
- | 34 | Maybe/Accept-part | aap_am |
142
- | 35 | Downplayer | t1 |
143
- | 36 | Self-talk | bd |
144
- | 37 | Tag-Question | ^g |
145
- | 38 | Declarative Wh-Question | qw^d |
146
- | 39 | Apology | fa |
147
- | 40 | Thanking | ft |
148
-
149
-
150
- ## Data Stats
151
-
152
- |Dialogue Act | Labels | Count | % | Train Count | Train % | Test Count | Test % | Val Count | Val %
153
- --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---:
154
- Statement-non-opinion | sd | 75136 | 37.62 | 72549 | 37.71 | 1317 | 32.30 | 1270 | 38.81
155
- Acknowledge (Backchannel) | b | 38281 | 19.17 | 36950 | 19.21 | 764 | 18.73 | 567 | 17.33
156
- Statement-opinion | sv | 26421 | 13.23 | 25087 | 13.04 | 718 | 17.61 | 616 | 18.83
157
- Uninterpretable | % | 15195 | 7.61 | 14597 | 7.59 | 349 | 8.56 | 249 | 7.61
158
- Agree/Accept | aa | 11123 | 5.57 | 10770 | 5.60 | 207 | 5.08 | 146 | 4.46
159
- Appreciation | ba | 4757 | 2.38 | 4619 | 2.40 | 76 | 1.86 | 62 | 1.89
160
- Yes-No-Question | qy | 4725 | 2.37 | 4594 | 2.39 | 84 | 2.06 | 47 | 1.44
161
- Yes Answers | ny | 3030 | 1.52 | 2918 | 1.52 | 73 | 1.79 | 39 | 1.19
162
- Conventional-closing | fc | 2581 | 1.29 | 2480 | 1.29 | 81 | 1.99 | 20 | 0.61
163
- Wh-Question | qw | 1976 | 0.99 | 1896 | 0.99 | 55 | 1.35 | 25 | 0.76
164
- No Answers | nn | 1374 | 0.69 | 1334 | 0.69 | 26 | 0.64 | 14 | 0.43
165
- Response Acknowledgement | bk | 1306 | 0.65 | 1271 | 0.66 | 28 | 0.69 | 7 | 0.21
166
- Hedge | h | 1226 | 0.61 | 1181 | 0.61 | 23 | 0.56 | 22 | 0.67
167
- Declarative Yes-No-Question | qy^d | 1218 | 0.61 | 1167 | 0.61 | 36 | 0.88 | 15 | 0.46
168
- Backchannel in Question Form | bh | 1053 | 0.53 | 1015 | 0.53 | 21 | 0.51 | 17 | 0.52
169
- Quotation | ^q | 983 | 0.49 | 931 | 0.48 | 17 | 0.42 | 35 | 1.07
170
- Summarize/Reformulate | bf | 952 | 0.48 | 905 | 0.47 | 23 | 0.56 | 24 | 0.73
171
- Other | fo_o_fw_"_by_bc | 879 | 0.44 | 857 | 0.45 | 15 | 0.37 | 7 | 0.21
172
- Affirmative Non-yes Answers | na | 847 | 0.42 | 831 | 0.43 | 10 | 0.25 | 6 | 0.18
173
- Action-directive | ad | 745 | 0.37 | 712 | 0.37 | 27 | 0.66 | 6 | 0.18
174
- Collaborative Completion | ^2 | 723 | 0.36 | 690 | 0.36 | 19 | 0.47 | 14 | 0.43
175
- Repeat-phrase | b^m | 687 | 0.34 | 655 | 0.34 | 21 | 0.51 | 11 | 0.34
176
- Open-Question | qo | 656 | 0.33 | 631 | 0.33 | 16 | 0.39 | 9 | 0.28
177
- Rhetorical-Question | qh | 575 | 0.29 | 554 | 0.29 | 12 | 0.29 | 9 | 0.28
178
- Hold Before Answer/Agreement | ^h | 556 | 0.28 | 539 | 0.28 | 7 | 0.17 | 10 | 0.31
179
- Reject | ar | 344 | 0.17 | 337 | 0.18 | 3 | 0.07 | 4 | 0.12
180
- Negative Non-no Answers | ng | 302 | 0.15 | 290 | 0.15 | 6 | 0.15 | 6 | 0.18
181
- Signal-non-understanding | br | 298 | 0.15 | 286 | 0.15 | 9 | 0.22 | 3 | 0.09
182
- Other Answers | no | 284 | 0.14 | 277 | 0.14 | 6 | 0.15 | 1 | 0.03
183
- Conventional-opening | fp | 225 | 0.11 | 220 | 0.11 | 5 | 0.12 | 0 | 0.00
184
- Or-Clause | qrr | 209 | 0.10 | 206 | 0.11 | 2 | 0.05 | 1 | 0.03
185
- Dispreferred Answers | arp_nd | 207 | 0.10 | 204 | 0.11 | 3 | 0.07 | 0 | 0.00
186
- 3rd-party-talk | t3 | 117 | 0.06 | 115 | 0.06 | 0 | 0.00 | 2 | 0.06
187
- Offers, Options Commits | oo_co_cc | 110 | 0.06 | 109 | 0.06 | 0 | 0.00 | 1 | 0.03
188
- Maybe/Accept-part | aap_am | 104 | 0.05 | 97 | 0.05 | 7 | 0.17 | 0 | 0.00
189
- Downplayer | t1 | 103 | 0.05 | 102 | 0.05 | 1 | 0.02 | 0 | 0.00
190
- Self-talk | bd | 103 | 0.05 | 100 | 0.05 | 1 | 0.02 | 2 | 0.06
191
- Tag-Question | ^g | 92 | 0.05 | 92 | 0.05 | 0 | 0.00 | 0 | 0.00
192
- Declarative Wh-Question | qw^d | 80 | 0.04 | 79 | 0.04 | 1 | 0.02 | 0 | 0.00
193
- Apology | fa | 79 | 0.04 | 76 | 0.04 | 2 | 0.05 | 1 | 0.03
194
- Thanking | ft | 78 | 0.04 | 67 | 0.03 | 7 | 0.17 | 4 | 0.12
195
-
196
-
197
- ![Label Frequencies](https://raw.githubusercontent.com/NathanDuran/Switchboard-Corpus/master/swda_data/metadata/Swda%20Label%20Frequency%20Distributions.png)
198
 
199
  ### Data Splits
200
 
201
- he data is split into the original training and test sets suggested by the authors (1115 training and 19 test). The remaining 21 dialogues have been used as a validation set.
 
 
 
 
 
 
 
 
 
202
 
203
  ## Dataset Creation
204
 
@@ -210,20 +193,7 @@ he data is split into the original training and test sets suggested by the autho
210
 
211
  #### Initial Data Collection and Normalization
212
 
213
- - Total number of utterances: 199740
214
- - Maximum utterance length: 133
215
- - Mean utterance length: 9.6
216
- - Total number of dialogues: 1155
217
- - Maximum dialogue length: 457
218
- - Mean dialogue length: 172.9
219
- - Vocabulary size: 22301
220
- - Number of labels: 41
221
- - Number of dialogue in train set: 1115
222
- - Maximum length of dialogue in train set: 457
223
- - Number of dialogue in test set: 19
224
- - Maximum length of dialogue in test set: 330
225
- - Number of dialogue in val set: 21
226
- - Maximum length of dialogue in val set: 299
227
 
228
  #### Who are the source language producers?
229
 
@@ -271,28 +241,32 @@ This work is licensed under a [Creative Commons Attribution-NonCommercial-ShareA
271
 
272
  ```
273
  @techreport{Jurafsky-etal:1997,
274
- Address = {Boulder, CO},
275
- Author = {Jurafsky, Daniel and Shriberg, Elizabeth and Biasca, Debra},
276
- Institution = {University of Colorado, Boulder Institute of Cognitive Science},
277
- Number = {97-02},
278
- Title = {Switchboard {SWBD}-{DAMSL} Shallow-Discourse-Function Annotation Coders Manual, Draft 13},
279
- Year = {1997}}
280
 
281
  @article{Shriberg-etal:1998,
282
- Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
283
- Journal = {Language and Speech},
284
- Number = {3--4},
285
- Pages = {439--487},
286
- Title = {Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?},
287
- Volume = {41},
288
- Year = {1998}}
289
 
290
  @article{Stolcke-etal:2000,
291
- Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
292
- Journal = {Computational Linguistics},
293
- Number = {3},
294
- Pages = {339--371},
295
- Title = {Dialogue Act Modeling for Automatic Tagging and Recognition of Conversational Speech},
296
- Volume = {26},
297
- Year = {2000}}
298
  ```
 
 
 
 
 
43
  - [Dataset Curators](#dataset-curators)
44
  - [Licensing Information](#licensing-information)
45
  - [Citation Information](#citation-information)
46
+ - [Contributions](#contributions)
47
 
48
  ## Dataset Description
49
 
50
  - **Homepage: [The Switchboard Dialog Act Corpus](http://compprag.christopherpotts.net/swda.html)**
51
+ - **Repository: [NathanDuran/Switchboard-Corpus](https://github.com/cgpotts/swda)**
52
  - **Paper:[The Switchboard Dialog Act Corpus](http://compprag.christopherpotts.net/swda.html)**
53
  = **Leaderboard: [Dialogue act classification](https://github.com/sebastianruder/NLP-progress/blob/master/english/dialogue.md#dialogue-act-classification)**
54
  - **Point of Contact: [Christopher Potts](https://web.stanford.edu/~cgpotts/)**
 
88
 
89
  An example from the dataset is:
90
 
91
+ `{'act_tag': 115, 'caller': 'A', 'conversation_no': 4325, 'damsl_act_tag': 26, 'from_caller': 1632, 'from_caller_birth_year': 1962, 'from_caller_dialect_area': 'WESTERN', 'from_caller_education': 2, 'from_caller_sex': 'FEMALE', 'length': 5, 'pos': 'Okay/UH ./.', 'prompt': 'FIND OUT WHAT CRITERIA THE OTHER CALLER WOULD USE IN SELECTING CHILD CARE SERVICES FOR A PRESCHOOLER. IS IT EASY OR DIFFICULT TO FIND SUCH CARE?', 'ptb_basename': '4/sw4325', 'ptb_treenumbers': '1', 'subutterance_index': 1, 'swda_filename': 'sw00utt/sw_0001_4325.utt', 'talk_day': '03/23/1992', 'text': 'Okay. /', 'to_caller': 1519, 'to_caller_birth_year': 1971, 'to_caller_dialect_area': 'SOUTH MIDLAND', 'to_caller_education': 1, 'to_caller_sex': 'FEMALE', 'topic_description': 'CHILD CARE', 'transcript_index': 0, 'trees': '(INTJ (UH Okay) (. .) (-DFL- E_S))', 'utterance_index': 1}`
 
 
92
 
93
  ### Data Fields
94
+
95
+ * `swda_filename`: (str) The filename: directory/basename.
96
+ * `ptb_basename`: (str) The Treebank filename: add ".pos" for POS and ".mrg" for trees
97
+ * `conversation_no`: (int) The conversation Id, to key into the metadata database.
98
+ * `transcript_index`: (int) The line number of this item in the transcript (counting only utt lines).
99
+ * `act_tag`: (list of str) The Dialog Act Tags (separated by ||| in the file). Check Dialog act annotations for more details.
100
+ * `damsl_act_tag`: (list of str) The Dialog Act Tags of the 217 variation tags.
101
+ * `caller`: (str) A, B, @A, @B, @@A, @@B
102
+ * `utterance_index`: (int) The encoded index of the utterance (the number in A.49, B.27, etc.)
103
+ * `subutterance_index`: (int) Utterances can be broken across line. This gives the internal position.
104
+ * `text`: (str) The text of the utterance
105
+ * `pos`: (str) The POS tagged version of the utterance, from PtbBasename+.pos
106
+ * `trees`: (str) The tree(s) containing this utterance (separated by ||| in the file). Use `[Tree.fromstring(t) for t in row_value.split("|||")]` to convert to (list of nltk.tree.Tree).
107
+ * `ptb_treenumbers`: (list of int) The tree numbers in the PtbBasename+.mrg
108
+ * `talk_day`: (str) Date of talk.
109
+ * `length`: (int) Length of talk in seconds.
110
+ * `topic_description`: (str) Short description of topic that's being discussed.
111
+ * `prompt`: (str) Long decription/query/instruction.
112
+ * `from_caller`: (int) The numerical Id of the from (A) caller.
113
+ * `from_caller_sex`: (str) MALE, FEMALE.
114
+ * `from_caller_education`: (int) Called education level 0, 1, 2, 3, 9.
115
+ * `from_caller_birth_year`: (int) Caller birth year YYYY.
116
+ * `from_caller_dialect_area`: (str) MIXED, NEW ENGLAND, NORTH MIDLAND, NORTHERN, NYC, SOUTH MIDLAND, SOUTHERN, UNK, WESTERN.
117
+ * `to_caller`: (int) The numerical Id of the to (B) caller.
118
+ * `to_caller_sex`: (str) MALE, FEMALE.
119
+ * `to_caller_education`: (int) Called education level 0, 1, 2, 3, 9.
120
+ * `to_caller_birth_year`: (int) Caller birth year YYYY.
121
+ * `to_caller_dialect_area`: (str) MIXED, NEW ENGLAND, NORTH MIDLAND, NORTHERN, NYC, SOUTH MIDLAND, SOUTHERN, UNK, WESTERN.
122
+
123
+
124
+ ### Dialog act annotations
125
+
126
+
127
+ | | name | act_tag | example | train_count | full_count |
128
+ |----- |------------------------------- |---------------- |-------------------------------------------------- |------------- |------------ |
129
+ | 1 | Statement-non-opinion | sd | Me, I'm in the legal department. | 72824 | 75145 |
130
+ | 2 | Acknowledge (Backchannel) | b | Uh-huh. | 37096 | 38298 |
131
+ | 3 | Statement-opinion | sv | I think it's great | 25197 | 26428 |
132
+ | 4 | Agree/Accept | aa | That's exactly it. | 10820 | 11133 |
133
+ | 5 | Abandoned or Turn-Exit | % | So, - | 10569 | 15550 |
134
+ | 6 | Appreciation | ba | I can imagine. | 4633 | 4765 |
135
+ | 7 | Yes-No-Question | qy | Do you have to have any special training? | 4624 | 4727 |
136
+ | 8 | Non-verbal | x | [Laughter], [Throat_clearing] | 3548 | 3630 |
137
+ | 9 | Yes answers | ny | Yes. | 2934 | 3034 |
138
+ | 10 | Conventional-closing | fc | Well, it's been nice talking to you. | 2486 | 2582 |
139
+ | 11 | Uninterpretable | % | But, uh, yeah | 2158 | 15550 |
140
+ | 12 | Wh-Question | qw | Well, how old are you? | 1911 | 1979 |
141
+ | 13 | No answers | nn | No. | 1340 | 1377 |
142
+ | 14 | Response Acknowledgement | bk | Oh, okay. | 1277 | 1306 |
143
+ | 15 | Hedge | h | I don't know if I'm making any sense or not. | 1182 | 1226 |
144
+ | 16 | Declarative Yes-No-Question | qy^d | So you can afford to get a house? | 1174 | 1219 |
145
+ | 17 | Other | fo_o_fw_by_bc | Well give me a break, you know. | 1074 | 883 |
146
+ | 18 | Backchannel in question form | bh | Is that right? | 1019 | 1053 |
147
+ | 19 | Quotation | ^q | You can't be pregnant and have cats | 934 | 983 |
148
+ | 20 | Summarize/reformulate | bf | Oh, you mean you switched schools for the kids. | 919 | 952 |
149
+ | 21 | Affirmative non-yes answers | na | It is. | 836 | 847 |
150
+ | 22 | Action-directive | ad | Why don't you go first | 719 | 746 |
151
+ | 23 | Collaborative Completion | ^2 | Who aren't contributing. | 699 | 723 |
152
+ | 24 | Repeat-phrase | b^m | Oh, fajitas | 660 | 688 |
153
+ | 25 | Open-Question | qo | How about you? | 632 | 656 |
154
+ | 26 | Rhetorical-Questions | qh | Who would steal a newspaper? | 557 | 575 |
155
+ | 27 | Hold before answer/agreement | ^h | I'm drawing a blank. | 540 | 556 |
156
+ | 28 | Reject | ar | Well, no | 338 | 346 |
157
+ | 29 | Negative non-no answers | ng | Uh, not a whole lot. | 292 | 302 |
158
+ | 30 | Signal-non-understanding | br | Excuse me? | 288 | 298 |
159
+ | 31 | Other answers | no | I don't know | 279 | 286 |
160
+ | 32 | Conventional-opening | fp | How are you? | 220 | 225 |
161
+ | 33 | Or-Clause | qrr | or is it more of a company? | 207 | 209 |
162
+ | 34 | Dispreferred answers | arp_nd | Well, not so much that. | 205 | 207 |
163
+ | 35 | 3rd-party-talk | t3 | My goodness, Diane, get down from there. | 115 | 117 |
164
+ | 36 | Offers, Options, Commits | oo_co_cc | I'll have to check that out | 109 | 110 |
165
+ | 37 | Self-talk | t1 | What's the word I'm looking for | 102 | 103 |
166
+ | 38 | Downplayer | bd | That's all right. | 100 | 103 |
167
+ | 39 | Maybe/Accept-part | aap_am | Something like that | 98 | 105 |
168
+ | 40 | Tag-Question | ^g | Right? | 93 | 92 |
169
+ | 41 | Declarative Wh-Question | qw^d | You are what kind of buff? | 80 | 80 |
170
+ | 42 | Apology | fa | I'm sorry. | 76 | 79 |
171
+ | 43 | Thanking | ft | Hey thanks a lot | 67 | 78 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
172
 
173
  ### Data Splits
174
 
175
+ I used info from the [Probabilistic-RNN-DA-Classifier](https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier) repo:
176
+ The same training and test splits as used by [Stolcke et al. (2000)](https://web.stanford.edu/~jurafsky/ws97).
177
+ The development set is a subset of the training set to speed up development and testing used in the paper [Probabilistic Word Association for Dialogue Act Classification with Recurrent Neural Networks](https://www.researchgate.net/publication/326640934_Probabilistic_Word_Association_for_Dialogue_Act_Classification_with_Recurrent_Neural_Networks_19th_International_Conference_EANN_2018_Bristol_UK_September_3-5_2018_Proceedings).
178
+
179
+ |Dataset |# Transcripts |# Utterances |
180
+ |-----------|:-------------:|:-------------:|
181
+ |Training |1115 |192,768 |
182
+ |Validation |21 |3,196 |
183
+ |Test |19 |4,088 |
184
+
185
 
186
  ## Dataset Creation
187
 
 
193
 
194
  #### Initial Data Collection and Normalization
195
 
196
+ The SwDA is not inherently linked to the Penn Treebank 3 parses of Switchboard, and it is far from straightforward to align the two resources Calhoun et al. 2010, §2.4. In addition, the SwDA is not distributed with the Switchboard's tables of metadata about the conversations and their participants.
 
 
 
 
 
 
 
 
 
 
 
 
 
197
 
198
  #### Who are the source language producers?
199
 
 
241
 
242
  ```
243
  @techreport{Jurafsky-etal:1997,
244
+ Address = {Boulder, CO},
245
+ Author = {Jurafsky, Daniel and Shriberg, Elizabeth and Biasca, Debra},
246
+ Institution = {University of Colorado, Boulder Institute of Cognitive Science},
247
+ Number = {97-02},
248
+ Title = {Switchboard {SWBD}-{DAMSL} Shallow-Discourse-Function Annotation Coders Manual, Draft 13},
249
+ Year = {1997}}
250
 
251
  @article{Shriberg-etal:1998,
252
+ Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
253
+ Journal = {Language and Speech},
254
+ Number = {3--4},
255
+ Pages = {439--487},
256
+ Title = {Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?},
257
+ Volume = {41},
258
+ Year = {1998}}
259
 
260
  @article{Stolcke-etal:2000,
261
+ Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
262
+ Journal = {Computational Linguistics},
263
+ Number = {3},
264
+ Pages = {339--371},
265
+ Title = {Dialogue Act Modeling for Automatic Tagging and Recognition of Conversational Speech},
266
+ Volume = {26},
267
+ Year = {2000}}
268
  ```
269
+
270
+ ### Contributions
271
+
272
+ Thanks to [@gmihaila](https://github.com/gmihaila) for adding this dataset.
dataset_infos.json CHANGED
@@ -1 +1 @@
1
- {"default": {"description": "The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2 with\nturn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information about the\nassociated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.\nThe SwDA is not inherently linked to the Penn Treebank 3 parses of Switchboard, and it is far from straightforward to\nalign the two resources. In addition, the SwDA is not distributed with the Switchboard's tables of metadata about the\nconversations and their participants.\n", "citation": "@techreport{Jurafsky-etal:1997,\n Address = {Boulder, CO},\n Author = {Jurafsky, Daniel and Shriberg, Elizabeth and Biasca, Debra},\n Institution = {University of Colorado, Boulder Institute of Cognitive Science},\n Number = {97-02},\n Title = {Switchboard {SWBD}-{DAMSL} Shallow-Discourse-Function Annotation Coders Manual, Draft 13},\n Year = {1997}}\n\n@article{Shriberg-etal:1998,\n Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},\n Journal = {Language and Speech},\n Number = {3--4},\n Pages = {439--487},\n Title = {Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?},\n Volume = {41},\n Year = {1998}}\n\n@article{Stolcke-etal:2000,\n Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},\n Journal = {Computational Linguistics},\n Number = {3},\n Pages = {339--371},\n Title = {Dialogue Act Modeling for Automatic Tagging and Recognition of Conversational Speech},\n Volume = {26},\n Year = {2000}}\n", "homepage": "http://compprag.christopherpotts.net/swda.html", "license": "Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License", "features": {"speaker": {"num_classes": 2, "names": ["A", "B"], "names_file": null, "id": null, "_type": "ClassLabel"}, "utterance_text": {"dtype": "string", "id": null, "_type": "Value"}, "dialogue_act_tag": {"num_classes": 41, "names": ["sd", "b", "sv", "%", "aa", "ba", "qy", "ny", "fc", "qw", "nn", "bk", "h", "qy^d", "bh", "^q", "bf", "fo_o_fw_\"_by_bc", "na", "ad", "^2", "b^m", "qo", "qh", "^h", "ar", "ng", "br", "no", "fp", "qrr", "arp_nd", "t3", "oo_co_cc", "aap_am", "t1", "bd", "^g", "qw^d", "fa", "ft"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "builder_name": "swda", "config_name": "default", "version": {"version_str": "0.0.0", "description": null, "major": 0, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 10912837, "num_examples": 192390, "dataset_name": "swda"}, "validation": {"name": "validation", "num_bytes": 189531, "num_examples": 3272, "dataset_name": "swda"}}, "download_checksums": {"https://github.com/NathanDuran/Switchboard-Corpus/raw/master/swda_data/train_set.txt": {"num_bytes": 8182921, "checksum": "0d4ca6fb17d985b8e0eb7ae7ca52bfdd9c6e7c4e9e07a8597d1c808937cb9e80"}, "https://github.com/NathanDuran/Switchboard-Corpus/raw/master/swda_data/val_set.txt": {"num_bytes": 143020, "checksum": "51be26976ad5aa76c01dcdcd85d9ecab216f88b019f5b7cca6afc07dd4879354"}, "https://github.com/NathanDuran/Switchboard-Corpus/raw/master/swda_data/test_set.txt": {"num_bytes": 168864, "checksum": "c54e677d887ab0efb5ef92250d61cca7a1e091b4eb556061dc6987d9e385cd72"}}, "download_size": 8494805, "post_processing_size": null, "dataset_size": 11102368, "size_in_bytes": 19597173}}
 
1
+ {"default": {"description": "The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2 with\nturn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information about the\nassociated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.\nThe SwDA is not inherently linked to the Penn Treebank 3 parses of Switchboard, and it is far from straightforward to\nalign the two resources. In addition, the SwDA is not distributed with the Switchboard's tables of metadata about the\nconversations and their participants.\n", "citation": "@techreport{Jurafsky-etal:1997,\n Address = {Boulder, CO},\n Author = {Jurafsky, Daniel and Shriberg, Elizabeth and Biasca, Debra},\n Institution = {University of Colorado, Boulder Institute of Cognitive Science},\n Number = {97-02},\n Title = {Switchboard {SWBD}-{DAMSL} Shallow-Discourse-Function Annotation Coders Manual, Draft 13},\n Year = {1997}}\n\n@article{Shriberg-etal:1998,\n Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},\n Journal = {Language and Speech},\n Number = {3--4},\n Pages = {439--487},\n Title = {Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?},\n Volume = {41},\n Year = {1998}}\n\n@article{Stolcke-etal:2000,\n Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},\n Journal = {Computational Linguistics},\n Number = {3},\n Pages = {339--371},\n Title = {Dialogue Act Modeling for Automatic Tagging and Recognition of Conversational Speech},\n Volume = {26},\n Year = {2000}}\n", "homepage": "http://compprag.christopherpotts.net/swda.html", "license": "Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License", "features": {"swda_filename": {"dtype": "string", "id": null, "_type": "Value"}, "ptb_basename": {"dtype": "string", "id": null, "_type": "Value"}, "conversation_no": {"dtype": "int64", "id": null, "_type": "Value"}, "transcript_index": {"dtype": "int64", "id": null, "_type": "Value"}, "act_tag": {"num_classes": 217, "names": ["b^m^r", "qw^r^t", "aa^h", "br^m", "fa^r", "aa,ar", "sd^e(^q)^r", "^2", "sd;qy^d", "oo", "bk^m", "aa^t", "cc^t", "qy^d^c", "qo^t", "ng^m", "qw^h", "qo^r", "aa", "qy^d^t", "qrr^d", "br^r", "fx", "sd,qy^g", "ny^e", "^h^t", "fc^m", "qw(^q)", "co", "o^t", "b^m^t", "qr^d", "qw^g", "ad(^q)", "qy(^q)", "na^r", "am^r", "qr^t", "ad^c", "qw^c", "bh^r", "h^t", "ft^m", "ba^r", "qw^d^t", "%", "t3", "nn", "bd", "h^m", "h^r", "sd^r", "qh^m", "^q^t", "sv^2", "ft", "ar^m", "qy^h", "sd^e^m", "qh^r", "cc", "fp^m", "ad", "qo", "na^m^t", "fo^c", "qy", "sv^e^r", "aap", "no", "aa^2", "sv(^q)", "sv^e", "nd", "\"", "bf^2", "bk", "fp", "nn^r^t", "fa^c", "ny^t", "ny^c^r", "qw", "qy^t", "b", "fo", "qw^r", "am", "bf^t", "^2^t", "b^2", "x", "fc", "qr", "no^t", "bk^t", "bd^r", "bf", "^2^g", "qh^c", "ny^c", "sd^e^r", "br", "fe", "by", "^2^r", "fc^r", "b^m", "sd,sv", "fa^t", "sv^m", "qrr", "^h^r", "na", "fp^r", "o", "h,sd", "t1^t", "nn^r", "cc^r", "sv^c", "co^t", "qy^r", "sv^r", "qy^d^h", "sd", "nn^e", "ny^r", "b^t", "ba^m", "ar", "bf^r", "sv", "bh^m", "qy^g^t", "qo^d^c", "qo^d", "nd^t", "aa^r", "sd^2", "sv;sd", "qy^c^r", "qw^m", "qy^g^r", "no^r", "qh(^q)", "sd;sv", "bf(^q)", "+", "qy^2", "qw^d", "qy^g", "qh^g", "nn^t", "ad^r", "oo^t", "co^c", "ng", "^q", "qw^d^c", "qrr^t", "^h", "aap^r", "bc^r", "sd^m", "bk^r", "qy^g^c", "qr(^q)", "ng^t", "arp", "h", "bh", "sd^c", "^g", "o^r", "qy^c", "sd^e", "fw", "ar^r", "qy^m", "bc", "sv^t", "aap^m", "sd;no", "ng^r", "bf^g", "sd^e^t", "o^c", "b^r", "b^m^g", "ba", "t1", "qy^d(^q)", "nn^m", "ny", "ba,fe", "aa^m", "qh", "na^m", "oo(^q)", "qw^t", "na^t", "qh^h", "qy^d^m", "ny^m", "fa", "qy^d", "fc^t", "sd(^q)", "qy^d^r", "bf^m", "sd(^q)^t", "ft^t", "^q^r", "sd^t", "sd(^q)^r", "ad^t"], "names_file": null, "id": null, "_type": "ClassLabel"}, "damsl_act_tag": {"num_classes": 43, "names": ["ad", "qo", "qy", "arp_nd", "sd", "h", "bh", "no", "^2", "^g", "ar", "aa", "sv", "bk", "fp", "qw", "b", "ba", "t1", "oo_co_cc", "+", "ny", "qw^d", "x", "qh", "fc", "fo_o_fw_\"_by_bc", "aap_am", "%", "bf", "t3", "nn", "bd", "ng", "^q", "br", "qy^d", "fa", "^h", "b^m", "ft", "qrr", "na"], "names_file": null, "id": null, "_type": "ClassLabel"}, "caller": {"dtype": "string", "id": null, "_type": "Value"}, "utterance_index": {"dtype": "int64", "id": null, "_type": "Value"}, "subutterance_index": {"dtype": "int64", "id": null, "_type": "Value"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "pos": {"dtype": "string", "id": null, "_type": "Value"}, "trees": {"dtype": "string", "id": null, "_type": "Value"}, "ptb_treenumbers": {"dtype": "string", "id": null, "_type": "Value"}, "talk_day": {"dtype": "string", "id": null, "_type": "Value"}, "length": {"dtype": "int64", "id": null, "_type": "Value"}, "topic_description": {"dtype": "string", "id": null, "_type": "Value"}, "prompt": {"dtype": "string", "id": null, "_type": "Value"}, "from_caller": {"dtype": "int64", "id": null, "_type": "Value"}, "from_caller_sex": {"dtype": "string", "id": null, "_type": "Value"}, "from_caller_education": {"dtype": "int64", "id": null, "_type": "Value"}, "from_caller_birth_year": {"dtype": "int64", "id": null, "_type": "Value"}, "from_caller_dialect_area": {"dtype": "string", "id": null, "_type": "Value"}, "to_caller": {"dtype": "int64", "id": null, "_type": "Value"}, "to_caller_sex": {"dtype": "string", "id": null, "_type": "Value"}, "to_caller_education": {"dtype": "int64", "id": null, "_type": "Value"}, "to_caller_birth_year": {"dtype": "int64", "id": null, "_type": "Value"}, "to_caller_dialect_area": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "swda", "config_name": "default", "version": {"version_str": "0.0.0", "description": null, "major": 0, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 128498512, "num_examples": 213543, "dataset_name": "swda"}, "validation": {"name": "validation", "num_bytes": 34749819, "num_examples": 56729, "dataset_name": "swda"}, "test": {"name": "test", "num_bytes": 2560127, "num_examples": 4514, "dataset_name": "swda"}}, "download_checksums": {"https://github.com/cgpotts/swda/raw/master/swda.zip": {"num_bytes": 14449197, "checksum": "0a08b8dd3992b446c8a920cacf7d246abe1e8a092a1ef7b5a2ed0352de9ccad2"}, "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/train_split.txt": {"num_bytes": 5574, "checksum": "b29fa1d73d1f4a0cb560844bf43084ce3507c35e11fc1fde940d946484643c8e"}, "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/dev_split.txt": {"num_bytes": 1499, "checksum": "557af363be3c6f56067660bc174bab8b9e7cdd1ab9bd39343b72098b32b05eda"}, "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/test_split.txt": {"num_bytes": 94, "checksum": "ce74a2ea11c5b1f7e585c527094f3bcd565d8586a3bea4015aeb8b43ddbec8b9"}}, "download_size": 14456364, "post_processing_size": null, "dataset_size": 165808458, "size_in_bytes": 180264822}}
dummy/0.0.0/dummy_data.zip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7666f8bc05cc699408411a0b0837c51006a6fb4a7c3df74c4608bee83d99fdfd
3
- size 1134
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:747ddbb2ac4accdc6d34dac5878936f9f0c7a0f614ad1e4d6ab58d97dfe49295
3
+ size 14841
swda.py CHANGED
@@ -17,10 +17,20 @@ Switchboard Dialog Act Corpus
17
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2,
18
  with turn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information
19
  about the associated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.
 
 
 
20
  """
21
 
22
  from __future__ import absolute_import, division, print_function
23
 
 
 
 
 
 
 
 
24
  import datasets
25
 
26
 
@@ -35,8 +45,7 @@ _CITATION = """\
35
  Year = {1997}}
36
 
37
  @article{Shriberg-etal:1998,
38
- Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, \
39
- Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
40
  Journal = {Language and Speech},
41
  Number = {3--4},
42
  Pages = {439--487},
@@ -45,8 +54,7 @@ _CITATION = """\
45
  Year = {1998}}
46
 
47
  @article{Stolcke-etal:2000,
48
- Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and \
49
- Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
50
  Journal = {Computational Linguistics},
51
  Number = {3},
52
  Pages = {339--371},
@@ -55,7 +63,6 @@ _CITATION = """\
55
  Year = {2000}}
56
  """
57
 
58
-
59
  # Description of dataset gathered from: https://github.com/cgpotts/swda#overview.
60
  _DESCRIPTION = """\
61
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2 with
@@ -73,23 +80,298 @@ _HOMEPAGE = "http://compprag.christopherpotts.net/swda.html"
73
  _LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License"
74
 
75
  # Dataset main url.
76
- _URL = "https://github.com/NathanDuran/Switchboard-Corpus/raw/master/swda_data/"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
77
 
78
 
79
  class Swda(datasets.GeneratorBasedBuilder):
80
  """
81
- Switchboard Dialog Act Corpus
 
 
82
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2,
83
  with turn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information
84
  about the associated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.
 
85
  """
86
 
87
- # Splits url extensions for train, validation and test.
88
- _URLS = {"train": _URL + "train_set.txt", "dev": _URL + "val_set.txt", "test": _URL + "test_set.txt"}
 
 
 
 
89
 
90
  def _info(self):
91
  """
92
- Specify the datasets.DatasetInfo object which contains informations and typings for the dataset.
93
  """
94
 
95
  return datasets.DatasetInfo(
@@ -98,60 +380,33 @@ class Swda(datasets.GeneratorBasedBuilder):
98
  # This defines the different columns of the dataset and their types.
99
  features=datasets.Features(
100
  {
101
- "speaker": datasets.ClassLabel(
102
- num_classes=2,
103
- names=[
104
- "A",
105
- "B",
106
- ],
107
- ),
108
- "utterance_text": datasets.Value("string"),
109
- "dialogue_act_tag": datasets.ClassLabel(
110
- num_classes=41,
111
- names=[
112
- "sd",
113
- "b",
114
- "sv",
115
- "%",
116
- "aa",
117
- "ba",
118
- "qy",
119
- "ny",
120
- "fc",
121
- "qw",
122
- "nn",
123
- "bk",
124
- "h",
125
- "qy^d",
126
- "bh",
127
- "^q",
128
- "bf",
129
- 'fo_o_fw_"_by_bc',
130
- "na",
131
- "ad",
132
- "^2",
133
- "b^m",
134
- "qo",
135
- "qh",
136
- "^h",
137
- "ar",
138
- "ng",
139
- "br",
140
- "no",
141
- "fp",
142
- "qrr",
143
- "arp_nd",
144
- "t3",
145
- "oo_co_cc",
146
- "aap_am",
147
- "t1",
148
- "bd",
149
- "^g",
150
- "qw^d",
151
- "fa",
152
- "ft",
153
- ],
154
- ),
155
  }
156
  ),
157
  supervised_keys=None,
@@ -167,37 +422,382 @@ class Swda(datasets.GeneratorBasedBuilder):
167
  """
168
  Returns SplitGenerators.
169
  This method is tasked with downloading/extracting the data and defining the splits.
 
 
 
 
 
 
 
 
170
  """
171
 
 
 
 
 
 
172
  urls_to_download = self._URLS
 
173
  downloaded_files = dl_manager.download_and_extract(urls_to_download)
174
 
175
  return [
 
 
 
 
 
176
  datasets.SplitGenerator(
177
- name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"], "split": "train"}
 
178
  ),
 
179
  datasets.SplitGenerator(
180
- name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"], "split": "validation"}
181
  ),
182
  ]
183
 
184
- def _generate_examples(self, filepath, split):
185
  """
186
  Yields examples.
187
  This method will receive as arguments the `gen_kwargs` defined in the previous `_split_generators` method.
188
  It is in charge of opening the given file and yielding (key, example) tuples from the dataset
189
  The key is not important, it's more here for legacy reason (legacy from tfds).
 
 
 
 
 
 
 
 
 
 
 
190
  """
191
 
192
- with open(filepath, encoding="utf-8") as f:
193
- for id_, row in enumerate(f):
 
 
 
 
 
 
 
194
 
195
- # Parse row into speaker info | utterance text | dialogue act tag.
196
- parsed_row = row.rstrip("\r\n").split("|")
197
 
198
- # Also returning dialogue act integer label.
199
- yield f"{split}-{id_}", {
200
- "speaker": parsed_row[0],
201
- "utterance_text": parsed_row[1],
202
- "dialogue_act_tag": parsed_row[2],
203
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2,
18
  with turn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information
19
  about the associated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.
20
+
21
+ This script is a modified version of the original swda.py from https://github.com/cgpotts/swda/blob/master/swda.py from
22
+ the original corpus repo. Modifications are made to accommodate the HuggingFace Dataset project format.
23
  """
24
 
25
  from __future__ import absolute_import, division, print_function
26
 
27
+ import csv
28
+ import datetime
29
+ import glob
30
+ import io
31
+ import os
32
+ import re
33
+
34
  import datasets
35
 
36
 
 
45
  Year = {1997}}
46
 
47
  @article{Shriberg-etal:1998,
48
+ Author = {Shriberg, Elizabeth and Bates, Rebecca and Taylor, Paul and Stolcke, Andreas and Jurafsky, Daniel and Ries, Klaus and Coccaro, Noah and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
 
49
  Journal = {Language and Speech},
50
  Number = {3--4},
51
  Pages = {439--487},
 
54
  Year = {1998}}
55
 
56
  @article{Stolcke-etal:2000,
57
+ Author = {Stolcke, Andreas and Ries, Klaus and Coccaro, Noah and Shriberg, Elizabeth and Bates, Rebecca and Jurafsky, Daniel and Taylor, Paul and Martin, Rachel and Meteer, Marie and Van Ess-Dykema, Carol},
 
58
  Journal = {Computational Linguistics},
59
  Number = {3},
60
  Pages = {339--371},
 
63
  Year = {2000}}
64
  """
65
 
 
66
  # Description of dataset gathered from: https://github.com/cgpotts/swda#overview.
67
  _DESCRIPTION = """\
68
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2 with
 
80
  _LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License"
81
 
82
  # Dataset main url.
83
+ _URL = "https://github.com/cgpotts/swda/raw/master/swda.zip"
84
+
85
+ # Dialogue act tags - long version 217 dialogue acts labels.
86
+ _ACT_TAGS = [
87
+ "b^m^r",
88
+ "qw^r^t",
89
+ "aa^h",
90
+ "br^m",
91
+ "fa^r",
92
+ "aa,ar",
93
+ "sd^e(^q)^r",
94
+ "^2",
95
+ "sd;qy^d",
96
+ "oo",
97
+ "bk^m",
98
+ "aa^t",
99
+ "cc^t",
100
+ "qy^d^c",
101
+ "qo^t",
102
+ "ng^m",
103
+ "qw^h",
104
+ "qo^r",
105
+ "aa",
106
+ "qy^d^t",
107
+ "qrr^d",
108
+ "br^r",
109
+ "fx",
110
+ "sd,qy^g",
111
+ "ny^e",
112
+ "^h^t",
113
+ "fc^m",
114
+ "qw(^q)",
115
+ "co",
116
+ "o^t",
117
+ "b^m^t",
118
+ "qr^d",
119
+ "qw^g",
120
+ "ad(^q)",
121
+ "qy(^q)",
122
+ "na^r",
123
+ "am^r",
124
+ "qr^t",
125
+ "ad^c",
126
+ "qw^c",
127
+ "bh^r",
128
+ "h^t",
129
+ "ft^m",
130
+ "ba^r",
131
+ "qw^d^t",
132
+ "%",
133
+ "t3",
134
+ "nn",
135
+ "bd",
136
+ "h^m",
137
+ "h^r",
138
+ "sd^r",
139
+ "qh^m",
140
+ "^q^t",
141
+ "sv^2",
142
+ "ft",
143
+ "ar^m",
144
+ "qy^h",
145
+ "sd^e^m",
146
+ "qh^r",
147
+ "cc",
148
+ "fp^m",
149
+ "ad",
150
+ "qo",
151
+ "na^m^t",
152
+ "fo^c",
153
+ "qy",
154
+ "sv^e^r",
155
+ "aap",
156
+ "no",
157
+ "aa^2",
158
+ "sv(^q)",
159
+ "sv^e",
160
+ "nd",
161
+ '"',
162
+ "bf^2",
163
+ "bk",
164
+ "fp",
165
+ "nn^r^t",
166
+ "fa^c",
167
+ "ny^t",
168
+ "ny^c^r",
169
+ "qw",
170
+ "qy^t",
171
+ "b",
172
+ "fo",
173
+ "qw^r",
174
+ "am",
175
+ "bf^t",
176
+ "^2^t",
177
+ "b^2",
178
+ "x",
179
+ "fc",
180
+ "qr",
181
+ "no^t",
182
+ "bk^t",
183
+ "bd^r",
184
+ "bf",
185
+ "^2^g",
186
+ "qh^c",
187
+ "ny^c",
188
+ "sd^e^r",
189
+ "br",
190
+ "fe",
191
+ "by",
192
+ "^2^r",
193
+ "fc^r",
194
+ "b^m",
195
+ "sd,sv",
196
+ "fa^t",
197
+ "sv^m",
198
+ "qrr",
199
+ "^h^r",
200
+ "na",
201
+ "fp^r",
202
+ "o",
203
+ "h,sd",
204
+ "t1^t",
205
+ "nn^r",
206
+ "cc^r",
207
+ "sv^c",
208
+ "co^t",
209
+ "qy^r",
210
+ "sv^r",
211
+ "qy^d^h",
212
+ "sd",
213
+ "nn^e",
214
+ "ny^r",
215
+ "b^t",
216
+ "ba^m",
217
+ "ar",
218
+ "bf^r",
219
+ "sv",
220
+ "bh^m",
221
+ "qy^g^t",
222
+ "qo^d^c",
223
+ "qo^d",
224
+ "nd^t",
225
+ "aa^r",
226
+ "sd^2",
227
+ "sv;sd",
228
+ "qy^c^r",
229
+ "qw^m",
230
+ "qy^g^r",
231
+ "no^r",
232
+ "qh(^q)",
233
+ "sd;sv",
234
+ "bf(^q)",
235
+ "+",
236
+ "qy^2",
237
+ "qw^d",
238
+ "qy^g",
239
+ "qh^g",
240
+ "nn^t",
241
+ "ad^r",
242
+ "oo^t",
243
+ "co^c",
244
+ "ng",
245
+ "^q",
246
+ "qw^d^c",
247
+ "qrr^t",
248
+ "^h",
249
+ "aap^r",
250
+ "bc^r",
251
+ "sd^m",
252
+ "bk^r",
253
+ "qy^g^c",
254
+ "qr(^q)",
255
+ "ng^t",
256
+ "arp",
257
+ "h",
258
+ "bh",
259
+ "sd^c",
260
+ "^g",
261
+ "o^r",
262
+ "qy^c",
263
+ "sd^e",
264
+ "fw",
265
+ "ar^r",
266
+ "qy^m",
267
+ "bc",
268
+ "sv^t",
269
+ "aap^m",
270
+ "sd;no",
271
+ "ng^r",
272
+ "bf^g",
273
+ "sd^e^t",
274
+ "o^c",
275
+ "b^r",
276
+ "b^m^g",
277
+ "ba",
278
+ "t1",
279
+ "qy^d(^q)",
280
+ "nn^m",
281
+ "ny",
282
+ "ba,fe",
283
+ "aa^m",
284
+ "qh",
285
+ "na^m",
286
+ "oo(^q)",
287
+ "qw^t",
288
+ "na^t",
289
+ "qh^h",
290
+ "qy^d^m",
291
+ "ny^m",
292
+ "fa",
293
+ "qy^d",
294
+ "fc^t",
295
+ "sd(^q)",
296
+ "qy^d^r",
297
+ "bf^m",
298
+ "sd(^q)^t",
299
+ "ft^t",
300
+ "^q^r",
301
+ "sd^t",
302
+ "sd(^q)^r",
303
+ "ad^t",
304
+ ]
305
+
306
+ # Damsl dialogue act tags version - short version 43 dialogue acts labels.
307
+ _DAMSL_ACT_TAGS = [
308
+ "ad",
309
+ "qo",
310
+ "qy",
311
+ "arp_nd",
312
+ "sd",
313
+ "h",
314
+ "bh",
315
+ "no",
316
+ "^2",
317
+ "^g",
318
+ "ar",
319
+ "aa",
320
+ "sv",
321
+ "bk",
322
+ "fp",
323
+ "qw",
324
+ "b",
325
+ "ba",
326
+ "t1",
327
+ "oo_co_cc",
328
+ "+",
329
+ "ny",
330
+ "qw^d",
331
+ "x",
332
+ "qh",
333
+ "fc",
334
+ 'fo_o_fw_"_by_bc',
335
+ "aap_am",
336
+ "%",
337
+ "bf",
338
+ "t3",
339
+ "nn",
340
+ "bd",
341
+ "ng",
342
+ "^q",
343
+ "br",
344
+ "qy^d",
345
+ "fa",
346
+ "^h",
347
+ "b^m",
348
+ "ft",
349
+ "qrr",
350
+ "na",
351
+ ]
352
 
353
 
354
  class Swda(datasets.GeneratorBasedBuilder):
355
  """
356
+ This is the HuggingFace Dataset class for swda.
357
+
358
+ Switchboard Dialog Act Corpus Hugging Face Dataset class.
359
  The Switchboard Dialog Act Corpus (SwDA) extends the Switchboard-1 Telephone Speech Corpus, Release 2,
360
  with turn/utterance-level dialog-act tags. The tags summarize syntactic, semantic, and pragmatic information
361
  about the associated turn. The SwDA project was undertaken at UC Boulder in the late 1990s.
362
+
363
  """
364
 
365
+ # Urls for each split train-dev-test.
366
+ _URLS = {
367
+ "train": "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/train_split.txt",
368
+ "dev": "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/dev_split.txt",
369
+ "test": "https://github.com/NathanDuran/Probabilistic-RNN-DA-Classifier/raw/master/data/test_split.txt",
370
+ }
371
 
372
  def _info(self):
373
  """
374
+ Specify the datasets.DatasetInfo object which contains information and typings for the dataset.
375
  """
376
 
377
  return datasets.DatasetInfo(
 
380
  # This defines the different columns of the dataset and their types.
381
  features=datasets.Features(
382
  {
383
+ "swda_filename": datasets.Value("string"),
384
+ "ptb_basename": datasets.Value("string"),
385
+ "conversation_no": datasets.Value("int64"),
386
+ "transcript_index": datasets.Value("int64"),
387
+ "act_tag": datasets.ClassLabel(num_classes=217, names=_ACT_TAGS),
388
+ "damsl_act_tag": datasets.ClassLabel(num_classes=43, names=_DAMSL_ACT_TAGS),
389
+ "caller": datasets.Value("string"),
390
+ "utterance_index": datasets.Value("int64"),
391
+ "subutterance_index": datasets.Value("int64"),
392
+ "text": datasets.Value("string"),
393
+ "pos": datasets.Value("string"),
394
+ "trees": datasets.Value("string"),
395
+ "ptb_treenumbers": datasets.Value("string"),
396
+ "talk_day": datasets.Value("string"),
397
+ "length": datasets.Value("int64"),
398
+ "topic_description": datasets.Value("string"),
399
+ "prompt": datasets.Value("string"),
400
+ "from_caller": datasets.Value("int64"),
401
+ "from_caller_sex": datasets.Value("string"),
402
+ "from_caller_education": datasets.Value("int64"),
403
+ "from_caller_birth_year": datasets.Value("int64"),
404
+ "from_caller_dialect_area": datasets.Value("string"),
405
+ "to_caller": datasets.Value("int64"),
406
+ "to_caller_sex": datasets.Value("string"),
407
+ "to_caller_education": datasets.Value("int64"),
408
+ "to_caller_birth_year": datasets.Value("int64"),
409
+ "to_caller_dialect_area": datasets.Value("string"),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
410
  }
411
  ),
412
  supervised_keys=None,
 
422
  """
423
  Returns SplitGenerators.
424
  This method is tasked with downloading/extracting the data and defining the splits.
425
+
426
+ Args:
427
+ dl_manager (:obj:`datasets.utils.download_manager.DownloadManager`):
428
+ Download manager to download and extract data files from urls.
429
+
430
+ Returns:
431
+ :obj:`list[str]`:
432
+ List of paths to data.
433
  """
434
 
435
+ # Download extract and return path of data file.
436
+ dl_dir = dl_manager.download_and_extract(_URL)
437
+ # Use swda/ folder.
438
+ data_dir = os.path.join(dl_dir, "swda")
439
+ # Handle partitions files.
440
  urls_to_download = self._URLS
441
+ # Download extract and return paths of split files.
442
  downloaded_files = dl_manager.download_and_extract(urls_to_download)
443
 
444
  return [
445
+ # Return whole data path and train splits file downloaded path.
446
+ datasets.SplitGenerator(
447
+ name=datasets.Split.TRAIN, gen_kwargs={"data_dir": data_dir, "split_file": downloaded_files["train"]}
448
+ ),
449
+ # Return whole data path and dev splits file downloaded path.
450
  datasets.SplitGenerator(
451
+ name=datasets.Split.VALIDATION,
452
+ gen_kwargs={"data_dir": data_dir, "split_file": downloaded_files["dev"]},
453
  ),
454
+ # Return whole data path and train splits file downloaded path.
455
  datasets.SplitGenerator(
456
+ name=datasets.Split.TEST, gen_kwargs={"data_dir": data_dir, "split_file": downloaded_files["test"]}
457
  ),
458
  ]
459
 
460
+ def _generate_examples(self, data_dir, split_file):
461
  """
462
  Yields examples.
463
  This method will receive as arguments the `gen_kwargs` defined in the previous `_split_generators` method.
464
  It is in charge of opening the given file and yielding (key, example) tuples from the dataset
465
  The key is not important, it's more here for legacy reason (legacy from tfds).
466
+
467
+ Args:
468
+ data_dir (:obj:`str`):
469
+ Path where is downloaded dataset.
470
+
471
+ split_file (:obj:`str`):
472
+ Path of split file used for train-dev-test.
473
+
474
+ Returns:
475
+ :obj:`list[str]`:
476
+ List of paths to data.
477
  """
478
 
479
+ # Read in the split file.
480
+ split_file = io.open(file=split_file, mode="r", encoding="utf-8").read().splitlines()
481
+ # Read in corpus data using split files.
482
+ corpus = CorpusReader(src_dirname=data_dir, split_file=split_file)
483
+ # Generate examples.
484
+ for i_trans, trans in enumerate(corpus.iter_transcripts()):
485
+ for i_utt, utt in enumerate(trans.utterances):
486
+ id_ = str(i_trans) + ":" + str(i_utt)
487
+ yield id_, {feature: utt[feature] for feature in self.info.features.keys()}
488
 
 
 
489
 
490
+ class CorpusReader:
491
+ """Class for reading in the corpus and iterating through its values."""
492
+
493
+ def __init__(self, src_dirname, split_file=None):
494
+ """
495
+ Reads in the data from `src_dirname` (should be the root of the
496
+ corpus). Assumes that the metadata file `swda-metadata.csv` is
497
+ in the main directory of the corpus, using that file to build
498
+ the `Metadata` object used throughout.
499
+
500
+ Args:
501
+ src_dirname (:obj:`str`):
502
+ Path where swda folder with all data.
503
+
504
+ split_file (:obj:`list[str`, `optional`):
505
+ List of file names used in a split (train, dev or test). This argument is optional and it will have a None value attributed inside the function.
506
+
507
+ """
508
+
509
+ self.src_dirname = src_dirname
510
+ metadata_filename = os.path.join(src_dirname, "swda-metadata.csv")
511
+ self.metadata = Metadata(metadata_filename)
512
+ self.split_file = split_file
513
+
514
+ def iter_transcripts(
515
+ self,
516
+ ):
517
+ """
518
+ Iterate through the transcripts.
519
+
520
+ Returns:
521
+ :obj:`Transcript`:
522
+ Transcript instance.
523
+ """
524
+
525
+ # All files names.
526
+ filenames = glob.glob(os.path.join(self.src_dirname, "sw*", "*.csv"))
527
+ # If no split files are mentioned just use all files.
528
+ self.split_file = filenames if self.split_file is None else self.split_file
529
+ # Filter out desired file names
530
+ filenames = [
531
+ file for file in filenames if os.path.basename(file).split("_")[-1].split(".")[0] in self.split_file
532
+ ]
533
+ for filename in sorted(filenames):
534
+ # Yield the Transcript instance:
535
+ yield Transcript(filename, self.metadata)
536
+
537
+ def iter_utterances(
538
+ self,
539
+ ):
540
+ """
541
+ Iterate through the utterances.
542
+
543
+ Returns:
544
+ :obj:`Transcript.utterances`:
545
+ Utterance instance object.
546
+ """
547
+
548
+ for trans in self.iter_transcripts():
549
+ for utt in trans.utterances:
550
+ # Yield the Utterance instance:
551
+ yield utt
552
+
553
+
554
+ class Metadata:
555
+ """
556
+ Basically an internal method for organizing the tables of metadata
557
+ from the original Switchboard transcripts and linking them with
558
+ the dialog acts.
559
+ """
560
+
561
+ def __init__(self, metadata_filename):
562
+ """
563
+ Turns the CSV file into a dictionary mapping Switchboard
564
+ conversation_no integers values to dictionaries of values. All
565
+ the keys correspond to the column names in the original
566
+ tables.
567
+
568
+ Args:
569
+ metadata_filename (:obj:`str`):
570
+ The CSV file swda-metadata.csv (should be in the main
571
+ folder of the swda directory).
572
+
573
+ """
574
+ self.metadata_filename = metadata_filename
575
+ self.metadata = {}
576
+ self.get_metadata()
577
+
578
+ def get_metadata(self):
579
+ """
580
+ Build the dictionary self.metadata mapping conversation_no to
581
+ dictionaries of values (str, int, or datatime, as
582
+ appropriate).
583
+ """
584
+
585
+ csvreader = csv.reader(open(self.metadata_filename))
586
+ header = next(csvreader)
587
+ for row in csvreader:
588
+ d = dict(list(zip(header, row)))
589
+ # Add integer number features to metadata.
590
+ for key in (
591
+ "conversation_no",
592
+ "length",
593
+ "from_caller",
594
+ "to_caller",
595
+ "from_caller_education",
596
+ "to_caller_education",
597
+ ):
598
+ d[key] = int(d[key])
599
+ talk_day = d["talk_day"]
600
+ talk_year = int("19" + talk_day[:2])
601
+ talk_month = int(talk_day[2:4])
602
+ talk_day = int(talk_day[4:])
603
+ # Make sure to convert date time to string to match PyArrow data formats.
604
+ d["talk_day"] = datetime.datetime(year=talk_year, month=talk_month, day=talk_day).strftime("%m/%d/%Y")
605
+ d["from_caller_birth_year"] = int(d["from_caller_birth_year"])
606
+ d["to_caller_birth_year"] = int(d["to_caller_birth_year"])
607
+ self.metadata[d["conversation_no"]] = d
608
+
609
+ def __getitem__(self, val):
610
+ """
611
+ Val should be a key in self.metadata; returns the
612
+ corresponding value.
613
+
614
+ Args:
615
+ val (:obj:`str`):
616
+ Key in self.metadata.
617
+
618
+ Returns:
619
+ :obj::
620
+ Corresponding value.
621
+
622
+ """
623
+
624
+ return self.metadata[val]
625
+
626
+
627
+ class Utterance:
628
+ """
629
+ The central object of interest. The attributes correspond to the
630
+ values of the class variable header:
631
+
632
+ """
633
+
634
+ # Metadata header file.
635
+ header = [
636
+ "swda_filename",
637
+ "ptb_basename",
638
+ "conversation_no",
639
+ "transcript_index",
640
+ "act_tag",
641
+ "caller",
642
+ "utterance_index",
643
+ "subutterance_index",
644
+ "text",
645
+ "pos",
646
+ "trees",
647
+ "ptb_treenumbers",
648
+ ]
649
+
650
+ def __init__(self, row, transcript_metadata):
651
+ """
652
+ Args:
653
+ row (:obj:`list`):
654
+ A row from one of the corpus CSV files.
655
+
656
+ transcript_metadata (:obj:`dict`):
657
+ A Metadata value based on the current `conversation_no`.
658
+
659
+ """
660
+
661
+ # Utterance data:
662
+ for i in range(len(Utterance.header)):
663
+ att_name = Utterance.header[i]
664
+ row_value = None
665
+ if i < len(row):
666
+ row_value = row[i].strip()
667
+ # Special handling of non-string values.
668
+ if att_name == "trees":
669
+ if row_value:
670
+ # Origianl code returned list of nltk.tree and used `[Tree.fromstring(t) for t in row_value.split("|||")]`.
671
+ # Since we're returning str we don't need to make any mondifications to row_value.
672
+ row_value = row_value
673
+ else:
674
+ row_value = "" # []
675
+ elif att_name == "ptb_treenumbers":
676
+ if row_value:
677
+ row_value = row_value # list(map(int, row_value.split("|||")))
678
+ else:
679
+ row_value = "" # []
680
+ elif att_name == "act_tag":
681
+ # I thought these conjoined tags were meant to be split.
682
+ # The docs suggest that they are single tags, thought,
683
+ # so skip this conditional and let it be treated as a str.
684
+ # row_value = re.split(r"\s*[,;]\s*", row_value)
685
+ # `` Transcription errors (typos, obvious mistranscriptions) are
686
+ # marked with a "*" after the discourse tag.''
687
+ # These are removed for this version.
688
+ row_value = row_value.replace("*", "")
689
+ elif att_name in ("conversation_no", "transcript_index", "utterance_index", "subutterance_index"):
690
+ row_value = int(row_value)
691
+ # Add attribute.
692
+ setattr(self, att_name, row_value)
693
+ # Make sure conversation number matches.
694
+ assert self.conversation_no == transcript_metadata["conversation_no"]
695
+ # Add rest of missing metadata
696
+ [setattr(self, key, value) for key, value in transcript_metadata.items()]
697
+ # Add damsl tags.
698
+ setattr(self, "damsl_act_tag", self.damsl_act_tag())
699
+
700
+ def __getitem__(self, feature):
701
+ """
702
+ Return utterance features as dictionary. It allows us to call an utterance object as a dictionary.
703
+ It contains same keys as attributes.
704
+
705
+ Args:
706
+ feature (:obj:`str`):
707
+ Feature value of utterance that is part of attributes.
708
+
709
+ Returns:
710
+ :obj:
711
+ Value of feature from utterance. Value type can vary.
712
+ """
713
+
714
+ return vars(self)[feature]
715
+
716
+ def damsl_act_tag(
717
+ self,
718
+ ):
719
+ """
720
+ Seeks to duplicate the tag simplification described at the
721
+ Coders' Manual: http://www.stanford.edu/~jurafsky/ws97/manual.august1.html
722
+ """
723
+
724
+ d_tags = []
725
+ tags = re.split(r"\s*[,;]\s*", self.act_tag)
726
+ for tag in tags:
727
+ if tag in ("qy^d", "qw^d", "b^m"):
728
+ pass
729
+ elif tag == "nn^e":
730
+ tag = "ng"
731
+ elif tag == "ny^e":
732
+ tag = "na"
733
+ else:
734
+ tag = re.sub(r"(.)\^.*", r"\1", tag)
735
+ tag = re.sub(r"[\(\)@*]", "", tag)
736
+ if tag in ("qr", "qy"):
737
+ tag = "qy"
738
+ elif tag in ("fe", "ba"):
739
+ tag = "ba"
740
+ elif tag in ("oo", "co", "cc"):
741
+ tag = "oo_co_cc"
742
+ elif tag in ("fx", "sv"):
743
+ tag = "sv"
744
+ elif tag in ("aap", "am"):
745
+ tag = "aap_am"
746
+ elif tag in ("arp", "nd"):
747
+ tag = "arp_nd"
748
+ elif tag in ("fo", "o", "fw", '"', "by", "bc"):
749
+ tag = 'fo_o_fw_"_by_bc'
750
+ d_tags.append(tag)
751
+ # Dan J says (p.c.) that it makes sense to take the first;
752
+ # there are only a handful of examples with 2 tags here.
753
+ return d_tags[0]
754
+
755
+
756
+ class Transcript:
757
+ """
758
+ Transcript instances are basically just containers for lists of
759
+ utterances and transcript-level metadata, accessible via
760
+ attributes.
761
+ """
762
+
763
+ def __init__(self, swda_filename, metadata):
764
+ """
765
+ Sets up all the attribute values:
766
+
767
+ Args:
768
+ swda_filename (:obj:`str`):
769
+ The filename for this transcript.
770
+
771
+ metadata (:obj:`str` or `Metadata`):
772
+ If a string, then assumed to be the metadata filename, and
773
+ the metadata is created from that filename. If a `Metadata`
774
+ object, then used as the needed metadata directly.
775
+
776
+ """
777
+
778
+ self.swda_filename = swda_filename
779
+ # If the supplied value is a filename:
780
+ if isinstance(metadata, str) or isinstance(metadata, str):
781
+ self.metadata = Metadata(metadata)
782
+ else: # Where the supplied value is already a Metadata object.
783
+ self.metadata = metadata
784
+ # Get the file rows:
785
+ rows = list(csv.reader(open(self.swda_filename, "rt")))
786
+ # Ge the header and remove it from the rows:
787
+ self.header = rows[0]
788
+ rows.pop(0)
789
+ # Extract the conversation_no to get the meta-data. Use the
790
+ # header for this in case the column ordering is ever changed:
791
+ row0dict = dict(list(zip(self.header, rows[1])))
792
+ self.conversation_no = int(row0dict["conversation_no"])
793
+ # The ptd filename in the right format for the current OS:
794
+ self.ptd_basename = os.sep.join(row0dict["ptb_basename"].split("/"))
795
+ # The dictionary of metadata for this transcript:
796
+ transcript_metadata = self.metadata[self.conversation_no]
797
+ for key, val in transcript_metadata.items():
798
+ setattr(self, key, transcript_metadata[key])
799
+ # Create the utterance list:
800
+ self.utterances = [Utterance(x, transcript_metadata) for x in rows]
801
+ # Coder's Manual: ``We also removed any line with a "@"
802
+ # (since @ marked slash-units with bad segmentation).''
803
+ self.utterances = [u for u in self.utterances if not re.search(r"[@]", u.act_tag)]