ParlaMint-GR

Subcorpora

Ελληνικό κοινοβουλευτικό σώμα κειμένου ParlaMint-GR

Source(s)

Minutes of the assembly of the greek parliament	2015-01-01 - 2022-02-01

Location(s)

Αθήνα

Ελλάδα

2015-01-01 - 2022-02-01

Names database

Transcriptions

Corpus search (59M tokens)

# Linguistically annotated corpus of parliamentary debates ParlaMint-GR

- Country: GR (Greece)
- Languages: el (Greek)
- Version: 4.1
- Handle: [http://hdl.handle.net/11356/1911](http://hdl.handle.net/11356/1911)

## Documentation

### Characteristics of the national parliament

The Greek parliament is a unicameral parliament with a multi-party political system. The proceedings are organized into Parliamentary terms (a term is the period between two general elections). Each Parliamentary term is divided in Sessions; a parliamentary term has regular Sessions, while extraordinary and special Sessions are also foreseen. Each Session is divided into Meetings (each meeting constitutes part of a parliamentary session), and each Meeting into Sittings (multiple sittings are possible, e.g., morning/afternoon sittings).

### Data source and acquisition

The ParlaMint-GR corpus contains the proceedings of the Greek Parliament Plenum meetings for the period 2015-01-01 - 2022-02-01. The proceedings source data (1,263 doc & docx files) were collected by scraping the Hellenic parliament official site
(https://www.hellenicparliament.gr/Praktika/Synedriaseis-Olomeleias).

The metadata for the members of parliament (MP) were obtained from various sources:

- Firstly, we got a list of all the MPs in tandem with the political party they belonged to during each period, by scraping the Hellenic parliament official site, where a dedicated page  (https://www.hellenicparliament.gr/Vouleftes/Diatelesantes-Vouleftes-Apo-Ti-Metapolitefsi-Os-Simera/) lists all Members of Parliament from 1974 until today, with their political affiliations.

In order to obtain the metadata for each government, i.e. the starting and ending date  of governance, the ministers of each government, their corresponding ministries with the relevant dates, as well as any resignations or suspensions, we scraped the Secretariat General for Legal and Parliamentary Affairs, a unit of the Office of the Prime Minister (https://gslegal.gov.gr/?page_id=776&sort=time), where this information is provided for all Greek governments since 1909.

- Regarding the acquisition of additional information about each speaker MP, such as their parliamentary roles (e.g. party presidents, parliament presidents and vice-presidents). and/or government positions, as well as their electoral districts, and their gender, we deployed the work of K. Dritsa 2020 (https://github.com/iMEdD-Lab/Greek_Parliament_Proceedings). In order to accommodate the project’s needs we made modifications and adjustments to the code.
- Additionally, for each political party we manually created the required metadata (Wikipedia page, official name, initials, party leader, foundation year, end of existence, coalitions, oppositions). Lastly, we manually edited cases where MPs changed parties or ceased to be part of political parties (due to death or suspension).
- Finally, in order to obtain an extensive list of male and female first and last names in all possible cases (Nominative, Genitive, Accusative, Vocative) in singular and plural number, we scraped the following Wiktionary sites:
   - https://el.wiktionary.org/wiki/Κατηγορία:Ανδρικά_ονόματα_(νέα_ελληνικά)
   - https://el.wiktionary.org/wiki/Κατηγορία:Γυναικεία_ονόματα_(νέα_ελληνικά)
   - https://el.wiktionary.org/wiki/Κατηγορία:Ανδρικά_επώνυμα_(νέα_ελληνικά)
   - https://el.wiktionary.org/wiki/Κατηγορία:Γυναικεία_επώνυμα_(νέα_ελληνικά

### Data encoding process

For each of the 1,263 proceedings’ files we made use of regexes and string matching rules in order to identify the beginning and the ending of each speech, the speaker and his/her role.

For each MP, a script was executed to automatically add the gender according to the previously constructed lists from Wiktionary.

To associate each speaker with the previously collected metadata (role, gender, political party and period) we used a script to find the Jaro-Winkler distance (https://en.wikipedia.org/wiki/Jaro%E2%80%93Winkler_distance) between the corresponding speaker and all the possible MPs that we had in our list. The similarity threshold was set at 0.95 to avoid false positives. This permitted us to match the names despite misspellings or ambiguities between the names. Again, the aforementioned GitHub repo was a big assist to this.

In order to produce the ParlaMint TEI xml encoded files we developed a java code that loads the aforementioned collection files and produces the required ParlaMint TEI xml outputs

Each utterance was annotated as being a proper speech by an MP or non-lexical vocal sounds by MPs (shouts, laughter, etc.) recorded in the minutes; these are characterised as Vocal by the TEI schema (https://clarin-eric.github.io/ParlaMint/#TEI.vocal). The distinction between Speech and Vocal was based on the name of the speaker, as recorded: if an utterance was recorded by the minutes not as belonging to a specific speaker, but instead the minutes assigned the utterance collectively to noun phrases such as [all MPs / many MPs / some MPs of party X] or used indefinite references such as [one MP from the House / one MP from party Y], then this utterance was encoded as a Vocal. Also, all empty utterances were identified as such and ignored during the TEI encoding process.

For each ParlaMint TEI xml file we kept track of the tags (text, body, div, u, seg, desc, vocal, s, w, pc, name, link, linkGrp, head, note, gap) that were used, in order to produce the cumulative descriptive statistics.

In particular each speech that was extracted from the Hellenic Parliament proceedings files was encoded as utterance <u> (https://clarin-eric.github.io/ParlaMint/#sec-uterrance) and each paragraph as <seg> (https://clarin-eric.github.io/ParlaMint/#TEI.seg).

For the linguistic annotated TEI xml files we parsed the output files of the ILSP NLP pipeline and encoded the identified sentences as <s> (https://clarin-eric.github.io/ParlaMint/#TEI.s), the words as <w> (https://clarin-eric.github.io/ParlaMint/#TEI.w), and the punctuations as <pc> (https://clarin-eric.github.io/ParlaMint/#TEI.pc). Finally, we needed to make use of the syntactic words format https://clarin-eric.github.io/ParlaMint/#sec-ana-norm for specific words (e.g. compound word constituting of the definite article prefixed by preposition, “στη” ,” στον”, “στους”, to the).

Finally, for the named entities and the dependency syntax we used the information from the linguistic annotation process; after parsing the output we were able to encode it using the tags <name> (https://clarin-eric.github.io/ParlaMint/#sec-ner), <linkGrp> and <link>
(https://clarin-eric.github.io/ParlaMint/#sec-parses).

### Corpus-specific metadata

There are no metadata specific to the Greek corpus.

### Structure

We strictly adhered to the ParlaMint schema.

### Linguistic annotation

For linguistic processing we used an NLP toolkit for Greek  (Prokopidis and Piperidis, 2020) developed at the Institute for Language and Speech Processing, and available through CLARIN:EL (ILSP Neural NLP Toolkit
 http://hdl.handle.net/11500/CLARIN-EL-0000-0000-67B2-3   The toolkit integrates modules, models and lexical resources for sentence splitting, tokenization, part of speech tagging, lemmatization, dependency parsing and named entity recognition. The output of the toolkit is in conllu format, extended with information about named entity spans (Person, Organization, Location and MISC NEs).

Prokopis Prokopidis and Stelios Piperidis. 2020. A Neural NLP toolkit for Greek. In 11th Hellenic Conference on Artificial Intelligence (SETN 2020).

Responsibles

Δημήτρης Γκούμας Βασίλης Παπαβασιλείου Προκόπης Προκοπίδης	TEI corpus encoding
Μαρία Γαβριηλίδου Στέλιος Πιπερίδης	Scientific supervisor