The CEPIC consists of parallel representation of speech segments in Cantonese, Putonghua and English. The following table shows the number of words (word token) and unique words (type) in each language.

Table 1. The composition of the CEPIC by language

The main speech types of CEPIC include the reading of government reports such as policy addresses and budget speeches, Q&A at press conferences, parliamentary debates, as well as remarks delivered at bilateral meetings. The following table shows the current composition of the corpus and some basic statistics of each subset.

Table 2. The composition of the CEPIC by speech types

The following is a list of words that have specific meaning in the CEPIC.

1. Annotated: It refers to a revised or newly transcribed version (when there are no readily available transcripts) of political speeches and their interpreted texts based on audios/videos collected from government websites and TV programme archives (also see Raw). The texts were annotated with prosodic and paralinguistic features that are of concern to the study of spoken language as well as interpreting.
2. CI: It refers to Consecutive Interpreting, a mode of interpreting employed in political settings when the interpreter speaks after the speaker finishes (part of) his/her speech (also see Interpreting Mode and SI).
3. Delivery Mode: It refers to the ways how a speech is delivered and includes mainly Monologue and Dialogue.
4. Dialogue: It refers to the Delivery Mode in which one or more political figures debate with or answer questions from the audience or each other.
5. Interpreter Language: It refers to the language that the interpreter interprets into.
6. Interpreting Mode: There are usually two modes of interpreting employed in political settings, i.e., CI and SI.
7. Location: It refers to the (capital) city where the political speech/interpreting is delivered.
8. Monologue: It refers to the Delivery Mode in which a political figure delivers the speech to the audience, with almost no interactions with the audience or other political figures.
9. Part of Speech (POS): It refers to the grammatical category a word belongs to, which is also called a word class. The CEPIC is POS tagged, i.e., each word is tagged with its part of speech. (Also refer to the section POS tagging for a full list of POS taggers used in the CEPIC).
10. Raw: It refers to the speech transcripts and their translations collected from government websites, with no annotations of prosodic or paralinguistic features (also see Annotated).
11. SI: It refers to Simultaneous Interpreting, a mode of interpreting employed in political settings when the interpreter speaks (usually in an interpreting booth) while the speaker deliveries his/her speech (also see CI and Interpreting Mode).
12. Speaker Language: It refers to the language that the speaker uses.
13. Speaker Role: It refers to the position of the speaker in a government.

The CEPIC is POS tagged with the assistance of Stanford CoreNLP 3.9.2 (Manning et al. 2014).

A semi-automatic process was employed to enhance the accuracy rate of machine tagging, in which all taggers were checked and revised based on subsets of manually checked testing data. Please click here(available soon) for a detailed account of the semi-automatic process employed in the POS tagging of CEPIC.

The following table provides a list of the POS taggers that appeared in the English subset of CEPIC, which is based on the Part-of-Speech Tagging Guidelines for the Penn Treebank Project (Santorini 1990, 6-7).

Table 1. POS taggers that appeared in the English subset of CEPIC (based on Santorini 1990: 6-7)

The following table provides a list of the POS taggers that appeared in the Chinese subset of CEPIC, which is based on the Part-Of-Speech Tagging Guidelines for the Penn Chinese Treebank (3.0) (Xia 2000: 37).

Table 2. POS taggers that appeared in the Chinese subset of CEPIC (based on Xia 2000: 37)

References:

Manning, Christopher D., Mihai Surdeanu, John Bauer, Jenny Finkel, Steven J. Bethard, and David McClosky. (2014). The Stanford CoreNLP Natural Language Processing Toolkit. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pp. 55-60.

Santorini, B. (1990). Part-Of-Speech Tagging Guidelines for the Penn Treebank Project (3rd revision, 2nd printing). Department of Linguistics, University of Pennsylvania.

Xia, Fei. (2000). The Part-Of-Speech Tagging Guidelines for the Penn Chinese Treebank (3.0). IRCS Technical Reports Series. 38.

Data of CEPIC were collected following specific steps and protocols. In particular, the “annotated” version of the CEPIC corpus was transcribed and annotated in a way that reflects features of spoken language data.

Speeches of CEPIC were manually revised or transcribed based on audios/videos with the speeches and their interpreting, if any. Apart from following a standardised process, the transcription of CEPIC aims to represent the spoken text as close as it was delivered. In addition, all Cantonese texts were transcribed in a way to capture spoken Cantonese features. Text and audio/video links were also included for those who may be interested in the sources of the speeches.

The following table shows the differences between the “raw” and “annotated” data.

As can be seen from the above table, the “annotated” version features annotations of different prosodic and paralinguistic features (e.g., pauses, fillers, repetitions and self-repair, etc.) that are of concern to the study of spoken language as well as interpreting.

Please click here(available soon) for a detailed account of the steps and protocols used in the transcription and annotation of CEPIC.