Data Collection

Overview

The data is politicians speaking to eachother about matters of concern of the Irish government.

https://www.oireachtas.ie/en/debates/

Dáil (House of Representatives)

174 TDs, Lower house, elected by public, most power.

Seanad (Senate)

Upper house.
Can debate, amend but not stop Dáil legislation.
The 60 members are chosen by panels of industry specific politicians (43), the Taoiseach (11) and university graduates (6).

Committee

Task groups of TDs and senators focused on specific topics (e.g. healthcare).

PQs (Parliamentary Questions)

Questions submitted and directed to specific people for answers.
Oral/Written but majority are written.
Oral PQs overlap with Dáil debates
Can filter written using URL/API to only select written PQs to avoid Dáil debate duplication.

Timeline

Data availibility

Dáil

1919 to Present

Seanad

1929 to Present

Committee

1924 to present

PQs

Jan 2025? to present

XML -> CSV (AI Generated Description)

This document describes how the Akoma Ntoso XML from the Oireachtas is transformed into the flat CSV debates_all.csv, and how to interpret each column—especially the text field which holds the raw debate content.

XML → CSV Mapping

Top‐level <debate type="…">
Each of the four debate types (Dáil, Seanad, Committee, Questions) becomes one set of rows.
Document‐level metadata
Extracted from <preface> and <FRBRWork>:
- doc_id ← /identification/FRBRWork/FRBRthis/@value
- date ← <debate date="…">
- title_ga, title_en ← <preface>/<block name="title_ga|en">/<docTitle>
- proponent_ga,proponent_en ← <preface>/<block name="proponent_ga|en">/<docProponent>
- status_ga, status_en ← <preface>/<block name="status_ga|en">/<docStatus>
- document_date ← <preface>/<block name="date_en">/<docDate>@date
- volume, number ← <preface>/<docNumber>@refersTo="#vol_…|#no_…"
Per‐element rows
Inside each <debateBody>, elements are normalized to rows:
- element_type = summary
  One row per <summary>: prelude notes, suspension, interruptions.
- element_type = speech
  One row per paragraph (<p>) in a <speech>: includes speaker id/name/role and timestamp.
- element_type = attendance (committees only)
  One row per committee sitting: the attendance column holds a semicolon‐joined list of all <person> names from the <rollCall>.
- element_type = question (questions only)
  One row per <question>: with separate question and written_answer columns and a combined text field.

CSV Columns Overview

• Common metadata:
doc_id, source_type, date, title_ga, title_en,
proponent_ga, proponent_en, status_ga, status_en,
document_date, volume, number, committee_name,
question_type, question_number.

• Structural context:
section_name, section_id, heading_text, heading_time.

• Element descriptors:
element_type, element_id.

• Speaker or person:
speaker_id, speaker_name, speaker_role, attendance.

• Timing:
recorded_time.

• Content:
- text ← Raw text of this row’s element.
- question ← The question text (only when element_type=question).
- written_answer ← The written answer text (only when element_type=question).

• Topic (questions only)
The to="…” attribute from <question>.

Understanding the text Field

The text column is your primary access to the debate content:

For summaries: shows the summary line exactly as in the XML.
For speeches: shows the full paragraph as spoken by a Deputy or Minister.
For attendance: repeats the roll‑call header (e.g. “Members present:”).
For question rows: concatenates the question and the written answer, so you can search Q&A in one go.

If you need to analyze just the spoken words, filter to element_type="speech" and read text.
For committee attendance, filter to element_type="attendance" and parse the semicolon‐separated attendance list.
For written questions/answers, filter to element_type="question" and use question vs. written_answer or the combined text.

Example Queries

All speeches by a given Deputy:

SELECT text
  FROM debates_all
 WHERE element_type = 'speech'
   AND speaker_name = 'Deputy Mary Lou McDonald';