Reykjavik University ASR drafts Althingi speech transcripts

AI system

System: Althingi ASR system (time-delay deep neural network acoustic model, n-gram language model, punctuation and paragraph models)
Model: Not assessed
Release: Not assessed
Application developer: Reykjavik University Language and Voice Lab (ASR for Althingi project)
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: About 6,300 hours of aligned parliamentary speech, of which a re-cleaned 1,000-hour subset trained the production recognizer (Interspeech 2019). The 542 hours of 2005–2016 speeches released as LDC2021S01 are a dataset publication, not the production training set. An existing manual transcription and editing workflow to slot into. Integration with Althingi's publishing systems from October 2018.
Human oversight: Not assessed
Relation to AI: The ASR produces the first-draft transcript. Althingi's editors still post-edit, so the benefit is editing time saved rather than full automation.
AI contribution: The ASR produces the first-draft transcript. Althingi's editors still post-edit, so the benefit is editing time saved rather than full automation.

Benefits

Outcome identifiers: Not assessed
Benefit: transcribers graded 77% of transcripts as Good
Category: Repetitive Task Automation
Status: Not assessed
Beneficiary groups: Organizations. Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Measurement

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: transcribers graded 77% of transcripts as Good
Comparison: Assumed alternative: the previous manual first-stage transcription by Althingi's transcribers
Results: Assumed comparison. Reference: First, a human transcriber listens to the audio and transcribes the speech. Result: the speech transcripts are now available sooner than before.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Money

Kind of value: Not assessed
Amount: Not assessed
Currency and prices: Not assessed
Period and unit: Not assessed
Realization: Not assessed
Recipients and payers: Not assessed
Gross or net: Not assessed
Included costs: Not assessed
Comparison: Not assessed
Method: Not assessed
Return and payback: Not assessed

People and places

Participants: Developer: Reykjavik University Language and Voice Lab (ASR for Althingi project). Operator: Alþingi (Icelandic parliament) speech department.
Jurisdictions: Benefit: Alþingi, Reykjavík, Iceland. Developer: Iceland. Operator: Iceland.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Follow-up

Conditions for replication: Editors accept ASR drafts (77% graded Good). Continued model maintenance by the developing lab.
Replication attempts: 1 follow-up entries. Benefit recurred: 1.
Later findings: Not assessed
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Evidence

Source numbers: AIBD-0035-source-0001–0005
Citations: Numbered references below.
Language: Not assessed
Source relationships: Developer, Independent
Original evidence: Not assessed
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Grades

Assessed benefits: Organizations. Society.
Depth: Organizations: 2. Society: 1.
Reasons: Organizations: Althingi replaced first-stage manual transcription with ASR drafts graded Good for 77% of speeches, with about 640 hours transcribed by November 2019. Society: Vísir reported on 4 March 2018 that the recognizer was released under an open-source license for Icelandic-language developers. A small, diffuse benefit.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently reported

Entry details

Kind

Not assessed

Related entries

Same task: automatic transcription of official proceedings for a public institution; Althingi's ASR drafts parliamentary transcripts, while the Italian judiciary's AI transcription recorded “illicit” for “licit”, harming the Genoa Prosecutor's Office and defendants.

Field

Government / language technology. Parliamentary transcription.

Purpose

Built specifically to transcribe Althingi speeches for post-editing and publication. This benefit was that purpose.

Dates

Date basis unassessed: 2019 (precision unassessed). Updated: 21 October 2026.

Stage

Not assessed

Version

SHA-256: 9e2fde75063b…

References

Other identifiers: AI Incident Database: 708.
Category mappings: Not assessed
Brief: Not assessed
Revisions: 7 revisions. Latest: 21 October 2026.

Sources

Peer-reviewed study Developer

Helgadóttir IR, Nikulásdóttir AB, Borský M, Fong JY, Kjaran R, Guðnason J (2019). The Althingi ASR System. Interspeech 2019, Graz.
Peer-reviewed study Developer

Helgadóttir IR et al. (2019). The Althingi ASR System. ISCA Archive abstract page.
Dataset Developer

Linguistic Data Consortium (2021). Althingi Parliamentary Speech, LDC2021S01 (authors Helgadóttir IR, Kjaran R, Nikulásdóttir AB, Guðnason J, the system's developers).
News report Independent author

Morgunblaðið (16 Nov 2019). Talgreinir skrifar ræður alþingismanna.
News report Independent author

Vísir (4 Mar 2018). Gefa út íslenskan talgreini með opnu hugbúnaðarleyfi.