Respeaking in the Danish Parliament’s Hansard

Introduction

During the last couple of decades, the Danish Parliament’s Hansard has taken a series of innovative technological steps forward in the production of the official report. One of the latest adaptations to new technology is the custom-built respeaking system Edixi, which allows the reporters to dictate instead of type the MPs’ speeches. In this article, I will describe why and how we use this system when writing the official report of the proceedings of the Danish Parliament.

From Pen to Microphone − From Paper to Bytes

In 1849, the Danish constitution, Grundloven, was signed by King Frederik VII and democracy in Denmark was born. The Danish Parliament was formed, and so came the need for an official report of its proceedings. For 175 years, reporters in the Office of the Folketing Hansard have made a verbatim official report of the proceedings. For more than the first 100 years, the reporters penned shorthand notes, which were then transformed into longhand before the text was set in lead letters and printed. In 1968, the parliamentary reporters took another technological leap forward and began recording the audio on tape to be played back when writing the report.

In 2007 we at Folketingstidende, the Office of the Folketing Hansard, jumped forward again when we started using a new custom-built speech recognition system called Edixi. The name Edixi is a contraction of “edit” and the Latin word “dixi”, which means “I have spoken”. We still use Edixi today. We respeak or dictate text instead of typing it while listening to the recording of the MPs’ speeches.

In 2009 we stopped printing the report on paper, publishing it only as HTML text on the main Parliament website www.folketinget.dk or as PDF files on the website of the Office of the Folketing Hansard, www.folketingstidende.dk.

Rendering an MP’s Speech into Text

Using an adapted version of the day’s agenda, a reporter is present in the plenary hall to make an electronic log of the proceedings, noting the speaking MP’s name and the item on the agenda. The reporter cuts the recording into speech segments and makes time stamps each time an MP starts speaking. This provides the reporter’s colleagues who write the report with a chronological list of empty speech segments and corresponding recordings to work on. The reporter in the plenary hall also makes notes of non-verbal communication and interjections, which are rare, as Danish MPs are very disciplined when debating.

The reporters writing the report then use this chronological list to reserve a sequence of segments—usually five to 10 minutes of speech—to transform into text. The list grows longer and longer as the proceedings go on, and when a reporter finishes their sequence they reserve the next available sequence.

The reporters’ finished sequences are published on the web right after the reporter clicks “Publish” in the text editor. An automatic style sheet converts the text in the editor into PDF and HTML files, and the system uploads them to the websites in a matter of minutes, forming the first or provisional edition. We aim to publish the complete report the next day, but during the busiest periods it may be delayed for a few days.The final, proof-read edition is published months later.   

Transformation of Soundwaves into Text

We use natural, human intelligence, not AI (yet), to transform the recordings into a slightly edited text with the correct punctuation, grammar and semantic structure. However, we accept some spoken language characteristics. We call the genre “written text with oral features”.

The reporters of the Danish Folketing Hansard have a substantial knowledge of Danish grammar, syntax and semantics, and have a pragmatic approach to transforming speech into text. There are “only” 179 MPs, so it is possible to get to know each MP’s background and idiolect: their specific way of speaking Danish. The most common procedure when we transcribe the report is to play back a sentence, listen to it carefully, quickly process it—deleting false starts, correcting the word order, and adding punctuation—and then dictate the text into the text editor. Sometimes, though, if an MP speaks clearly and with the correct semantics we may be able to dictate the text while listening to it, more or less like simultaneous interpreters—although, unlike interpreters, we can pause the recording.

A Uniform, Substantially Verbatim Report

The report is substantially verbatim. We edit it slightly in correspondence with a few general editing guidelines set up by the Presidium of the Danish Parliament. The spoken language is adapted into a colloquial and syntactically coherent text with a somewhat liberal approach to the conventions of written language. The speeches are edited with great care to ensure that the speaker’s intentions are clear. Factual errors and slips of the tongue are corrected, unless they are noticed and commented on by another MP.

We aim to edit the report as uniformly as possible. This is done by having joint training sessions when Parliament is not sitting and by discussing the difficulties we encounter in the editing process with each other. We also find many answers to questions on grammar, spelling, punctuation and the formal style of the report in our comprehensive online work of reference or style guide called Sprogbasen. You can read about this style guide in more detail in another article by Michael Ejstrup in this issue of [LINK].   

MPs do not need to approve the official report of their speeches before it is published but they can ask for corrections if they feel they have been misquoted. This happens only very rarely.

Why Not Use Speaker Independent Automatic Speech Recognition?

More and more people are using speaker independent automatic speech recognition (ASR) so why don’t we use it in the Office of the Folketing Hansard? This is because we are very pleased with our respeaking system, Edixi, with which we are familiar and comfortable. It simply works very well, and the quality of the recognition is very good; approximately 95% to 97% of the words are correct. The final editing is swift because the text we have dictated already has punctuation and paragraphs, and has been edited in the dictation process.

However, we visited both Stortinget, the Norwegian Parliament, and Eduskunta, the Finnish Parliament, in the autumn of 2023 and saw how they have begun to use speaker independent ASR successfully in their reporting. It is quite possible that the Danish Parliament will also consider an ASR solution like those in the near future.

Lars Busch Nielsen is a reporter in the Danish Parliament.