Processing the HCRC Map Task Corpus

Utilities for Processing the HCRC Map Task Corpus for the purpose of dialogue act (DA) classification. The data has been randomly split, with the training set comprising 80% of the dialogues (102), and test and validation sets 10% each (13).

Scripts

The maptask_to_text.py script processes all dialogues into a plain text format. Individual dialogues are saved into directories corresponding to the set they belong to (train, test, etc). All utterances in a particular set are also saved to a text file.

The utilities.py script contains various helper functions for loading/saving the data.

The process_transcript.py includes functions for processing each dialogue.

The maptask_metadata.py generates various metadata from the processed dialogues and saves them as a dictionary to a pickle file. The words, labels and frequencies are also saved as plain text files in the /metadata directory.

Data Format

Utterance are tagged with the Maptask Coding Scheme for DA.

By default:

• Utterances are written one per line in the format Speaker | Utterance Text | Dialogue Act Tag.
• Setting the utterance_only_flag == True, will change the default output to only one utterance per line i.e. no speaker or DA tags.
• Utterances marked as Uncodable ('uncodable' tag) are removed.
• Incomplete words have been removed i.e. 'th--'.

Example Utterances

g|okay|ready

g|starting off we are above a caravan park|instruct

f|mmhmm|acknowledge

Dialogue Acts

Dialogue Act	Labels	Count	%	Train Count	Train %	Test Count	Test %	Val Count	Val %
Acknowledge	acknowledge	5605	20.94	4433	21.04	527	20.29	645	20.82
Instruct	instruct	4267	15.94	3390	16.09	417	16.06	460	14.85
Yes-Reply	reply_y	3230	12.07	2530	12.01	304	11.71	396	12.78
Explain	explain	2160	8.07	1669	7.92	219	8.43	272	8.78
Check	check	2137	7.99	1683	7.99	232	8.93	222	7.17
Ready	ready	2062	7.70	1559	7.40	222	8.55	281	9.07
Check Attention	align	1778	6.64	1444	6.85	130	5.01	204	6.58
Yes-No-Question	query_yn	1758	6.57	1350	6.41	191	7.35	217	7.00
Clarify	clarify	1193	4.46	970	4.60	116	4.47	107	3.45
Non Yes-No-Reply	reply_w	916	3.42	729	3.46	83	3.20	104	3.36
No-Reply	reply_n	884	3.30	692	3.28	101	3.89	91	2.94
Non Yes-No-Question	query_w	772	2.88	618	2.93	55	2.12	99	3.20

Metadata

• Total number of utterances: 26743
• Max utterance length: 115
• Mean utterance length: 6.15
• Total Number of dialogues: 128
• Max dialogue length: 682
• Mean dialogue length: 208.93
• Vocabulary size: 1797
• Number of labels:12
• Number of speakers: 2

Train set

• Number of dialogues: 102
• Max dialogue length: 682
• Mean dialogue length: 206.39
• Number of utterances: 21052

Test set

• Number of dialogues: 13
• Max dialogue length: 314
• Mean dialogue length: 212.46
• Number of utterances: 2762

Val set

• Number of dialogues: 13
• Max dialogue length: 439
• Mean dialogue length: 225.31
• Number of utterances: 2929

Keys and values for the metadata dictionary

• num_utterances = Total number of utterance in the full corpus.
• max_utterance_len = Number of words in the longest utterance in the corpus.
• mean_utterance_len = Average number of words in utterances.
• num_dialogues = Total number of dialogues in the corpus.
• max_dialogues_len = Number of utterances in the longest dialogue in the corpus.
• mean_dialogues_len = Average number of utterances in dialogues.
• word_freq = Dataframe with Word and Count columns.
• vocabulary = List of all words in vocabulary.
• vocabulary_size = Number of words in the vocabulary.
• label_freq = Dataframe containing all data in the Dialogue Acts table above.
• labels = List of all DA labels.
• num_labels = Number of labels used from the Maptask data.
• speakers = List of all speakers.
• num_speakers = Number of speakers in the Maptask data.

Each data set also has:

• <setname>_num_utterances = Number of utterances in the set.
• <setname>_num_dialogues = Number of dialogues in the set.
• <setname>_max_dialogue_len = Length of the longest dialogue in the set.
• <setname>_mean_dialogue_len = Mean length of dialogues in the set.