For the past few years, professional discussion in our field has been dominated by the rise of AI, particularly automatic speech recognition, ASR. They have revolutionised reporting and transcription by providing unforeseen efficiency and productivity. Whole pages of text are produced automatically in just seconds. The results are becoming more accurate each year, and a properly taught AI tool when compared to the work of human professionals might even seem to edit the text with as much expertise.
But AI is not an expert. It is a big calculator that is counting probabilities and then acting based on them. With massive training data, it can succeed in following the motions in a fairly straightforward task like the physical side of transforming spoken words into written ones. But however useful and valuable it is, it cannot really make strategic, value-based decisions for us.
For example, the official parliamentary report has several potential audiences. These include the voters and the media, MPs and ministers, and researchers from various fields. AI might identify them based on prior literature on the topic, but it is for human professionals to decide what order they should be put in when making practical reporting decisions.
Voters or the general public are a very diverse category, for instance. Often, when I talk to people about parliamentary speeches, it surprises me how people find different things important in them. The same goes for the media. Just when I think that I have its interests figured out, a big news story is published where the focus is on something surprising, such as on a dialect that an MP used, or the type of shoes that they wore in the debate. Should elements like these be addressed in the official report? Why? How? To what extent?
MPs and ministers are also often mentioned as important recipients of the parliamentary report. That is easy to agree on, but how should that appear in practice? If the interests of MPs and ministers are in conflict with those of the voters or media, which should be prioritised? For example, an MP might want to correct the official report if they uttered the wrong name of a poet in a cultural debate because it was just a slip of the tongue. However, some people might find the mistake telling and important to know. Whose expectations should be met?
Researchers are also a diverse group. Some historians and political scientists might be satisfied with heavy editing as long as the content of the speeches has been preserved intact. On the other hand, some linguists and social scientists would like the report to be almost without any editing. This kind of report might be very authentic, but would it serve its purpose for other readers? Where should the line be drawn?
There is probably not a single right answer to these questions. Whatever it is for us in each case, it should be determined carefully. It is not a choice that should be left to a machine.
This spring issue is filled with a wide array of articles on making professional reports or transcripts for different audiences. Carlo Eugeni begins by presenting an idea for a universal report. In this model, reporting professionals use improving technology and increasing automation to create a flexible, personalised report to better serve the needs of different types of recipients. Daniela Eichmeyer-Hell & Anja Rau write about user-centered speech-to-text interpreting. In their article, they suggest several tactics with which the interpreter can enhance the readability of the text so that it best serves the client’s needs. Henk-Jan Eras, in turn, describes special reporting services at the Dutch Parliament, such as live captioning and sign language interpreting, that have been specifically targeted at people who are hard of hearing. He also describes their use of customer panels that have provided them with essential information about what their target groups actually want.
To serve different audiences optimally, the practices and workflows of reporting must be organised appropriately. Allowing for new working methods that are both efficient and supportive of staff well-being, new AI-based reporting tools have shown much promise around the world. Louis Peckstadt, Katrien Van Mulders and Lieve Beullens show how AI-supported parliamentary reporting has been implemented in the Flemish Parliament. Julia Schöllauf, Bettina Brixa and Dario Summer present the new in-house speech-to-text system that has been developed for parliamentary reporters in the Austrian parliament. Both articles report great benefits in taking up the new, AI-based reporting system, even though it also presents some new challenges to solve.
After the first excitement about AI solutions, it has become more and more evident that the centre stage still belongs to trained professionals who operate the tools and supervise their results. D’Arcy McPherson from the Legislative Assembly of British Columbia, Canada, presents how their reporting team renewed their work process after taking up AI-based reporting tools, so that they could strengthen the resilience of their workflow and make the best possible use of their professional reporters. Ana Rita Pereira and Paulo Granja emphasise that AI is not just a technical tool, showing that it might have a considerable impact on the linguistic and editorial choices that are made in reporting.
Tiro continues to provide a forum for active discussion in the field of reporting and transcription. Joanna Lipkowska gives an insightful conference report on the recent seminar of the Commonwealth Hansard Editors Association, while Luz Belenguer Cortés has produced a thorough book review of a recent professional book on “capturing talk”. In his scientific column, Carlo Eugeni reminds us that in professional reporting capturing talk is sometimes not enough. The speech event also consists of various non-verbal elements, such as intonation and body language, which can be essential to the report. Their meaning and significance can also vary – particularly when we ask who is it all for.
Eero Voutilainen is Tiro’s editor-in-chief.
