Edition 10 – Triinu Hiob

Triinu, thanks for joining me today and for giving us an inside on recording and transcribing of hearings. I had the pleasure of listening to your presentation during the Baltic Arbitration Days, but not everyone was as lucky. So, I am excited that you agreed to share your thoughts for this Digital Coffee Break in Arbitration. Before we go into detail, could you briefly explain the differences between recording and transcribing a hearing?

Thank you so much for inviting, Svenja!

The aim of both recording and transcribing a hearing is the same: to preserve information. Recording generates an accurate reproduction of what happened at the hearing, either as an audio or a video file. Transcription generates a text file and is already a conversion.

What are the pros and cons between recording and transcribing a hearing?

As mentioned, recording enables to have a very precise reproduction of the hearing – there are no mistakes and it is possible to hear not only the text but also the tone and eventual changes in it, in case of video recording also see the body language, etc. So, recordings tend to preserve more information than transcripts. However, since arbitral awards are delivered as text files, it is more convenient to work with a text file transcript when writing an award.

When transcribing a hearing, do you prefer working with an actual person or do you use a program?

I tend not to add the extra cost of having an actual person transcribing the hearing, except perhaps myself as the case may be.

I imagine that artificial intelligence might help tremendously with transcribing a hearing. What are the benefits of using a program compared to an actual court reporter?

The major benefit of using a program is the cost and rapidity ratio. The programs enable to have a transcript very quickly at a fraction of the cost that is involved in using an actual real-time court reporter. Hours of hearing are converted to a transcript within just a few dozen of minutes.

One relevant factor in each arbitration indeed are the costs. How expensive are recordings/transcriptions, especially comparing the cost for using a program compared to an actual court reporter? 

Recording is very easy and low cost these days. Dictaphones have been used for dozens of years. Today each smart device has the recording function. And in these strange times when we have moved a lot of our meetings and hearings to Teams, Zoom, Skype Business and other similar platforms, recording just takes one click.

The cost of transcribing depends on whether only a program or a combination of a program and some human help is used. Many service providers offer, in addition to the automatic transcript generated by a computer program, the opportunity of the transcript being edited by an actual person. This of course adds some extra cost. But if we speak about the created by a program transcript only, audio or video converted to text, then the cost for one hearing day may often be even a two-digit amount.

If you can get an audio/video file converted into a text for less than EUR 100, are there any scenarios – in your opinion – that would justify the extra costs for a court reporter?

If the case involves a lot of witness evidence, it is easier to work with a court reporter than edit an automatically created transcript. Plus when it comes to cost, it is always a question of proportions. In high value disputes the cost of using a court reporter may be well proportionate with what is at stake and what the overall costs are. But in small cases one needs to keep the costs low.

There are several service providers who offer artificial intelligence. What are the key points when deciding which one to use?

I would say that the most crucial point is the data security. Is the communication encrypted? Where is the service provider based and what is the data protection level in that country? Losing information in arbitration or falling thereof into the wrong hands is every party’s and tribunal’s worst nightmare.

Considering the low cost and rapidity of automatic transcribing, it seems like an ideal solution for hearings. Or are there any setbacks?

At least today, it is not ideal (yet). Computer programs do not think, they just write. To have a truly proper transcript one needs to work on it after receiving the automatically generated text file, to edit the text.

You just mentioned that you have to work on the text, after receiving the transcript; how much work is involved to go through the documents and to eliminate mistakes such as wrongly written names etc.?

The work volume after the receipt of an automatically generated transcript depends on a number of things, such as people’s eventual accents, use of filler words, foreign, non-English or otherwise unusual names, etc. It is my experience that mistakes in foreign names are common, so the automatic transcript needs to be edited there. What is trickier is if a speaker takes many small breaks to think, humming “uh” or “um”. The machine does not differ between the important and unimportant and diligently reflects all of these. This sometimes makes the text very inconvenient to read and one has to invest hours of work to edit the text.

What happens if the verbatim record – for whatever reason – misses a word that might be crucial? I can imagine that some people speak rather quickly and not entirely clear and all of the sudden the “didn’t” comes out as a “did”. Is there an assumption that the verbatim record is generally correct?

If there has been audio or video recording at the hearing, then to me it seems hopeless to argue that what can be heard in the audio file or seen in the video is not correct. If a word is not audible in the recording, it probably wasn’t audible at the hearing either.

As regards to a transcript, I would not dare to say that this always counts as correct, notwithstanding whether the transcript has been written by a human being or generated by a computer program. Both of them can make mistakes. But of course it is a question to put in front of the parties in the beginning of the proceedings or before the hearing: what weight they would be willing to render to the transcript? Are they willing to assume that the transcripts are correct, or should the recording prevail?

How do you ensure that what has been recorded and put into a protocol is correct and minimize the scenarios as mentioned in the previous question?

I prefer to have both, the recording in audio or video format, and the transcript as a text. If there is an argument that something is wrong in the transcript, we can go back to the recording that reproduces what happened at the hearing, enables us to relive the hearing.

Have you ever received a transcript that was not useable or that the recording did not take place?

I have not experienced that the recording did not take place, but I have had an automatically generated transcript that in parts was hardly readable. It was so because one speaker both had a strong accent and had not prepared well for the hearing, so they took many “uh” breaks, changed their thought midway of sentences so that there were both repetitions and missing words in the sentences. Somehow if listening to the speech directly, seeing the speaker in person or at least on screen, one can understand what is said. This is probably due to our human nature: experience, empathy, adaptability. But if the machine transcribes the text word by word, the transcript may be unreadable.

What are the consequences when due to a technical error, the transcript is not usable but the parties have agreed to transcribe the hearing?

If it is truly a technical error in generating the transcript and the hearing has been recorded as well, then it is pretty easy to generate a new transcript from the recording. I would say that the parties’ agreement to transcribe the hearing can be extended to the replacement of the transcript that is not usable due to a technical error.

If the transcript does exist but is partly unreadable due to a speaker’s way of speaking, it is not a technical issue. Then I would say that it is the risk the respective party bears.

If the parties have agreed to transcribe the hearing and to use a certain technical solution for this, and one’s presentation lacks quality insofar that it cannot be followed in the transcript, the party has not used their opportunity to properly present their case.

Do you remember the first time using a recorded audio or video file?

The first time I worked with a recorded audio file only, without transcript, it took me by surprise how much time I spent to listen the file again and again and to search the relevant parts. I had taken notes at the hearing – that is what I absolutely always do, notwithstanding of recording, transcribing, etc. – but I went back to the recording for the precise original wording. And though I enjoyed the opportunity to relive the hearing, it felt a bit time-consuming to cite the text in a document.

You are from Estonia, a country that is known as technologically well developed. Do local hearing recording and transcribing technology exist, for instance for court proceedings, that could also be used in arbitration?

By now all the court hearings in Estonia are recorded, and they are transcribed as well. However, as regards to the automation, there is still a way to go. Estonian courts used to use the Liberty Digital Court Recorder that is, to my knowledge, developed in the U.S. By now there is a recording software developed locally, called Salme. This enables the recorded audio file to be integrated within the transcript, which is a document file, so that it is possible to listen to the relevant part of the recording directly from the transcript, by clicking that part. For a party representative these transcripts are relatively convenient and easy to use. However, there is currently no artificial intelligence involved in generating the transcript, for the court secretary it is all manual work, both preparing the transcript and adding the links of the recording. To my knowledge, the work with creating a functionality of automatic transcription is going on but it is not yet ready.

Another question is that, at least so far, this technical solution is available to courts only and the state does not offer it for private use.

Where will we stand in 5, 10 or even 20 years from now?

I believe that we are moving towards combining different formats of presenting information. Should an arbitral award be text only? We already see parts of documentary evidence copied in the awards as picture files. Why not also insert audio or video files of witness testimonies, instead of citing these? This is technically possible already today; we just have to get used to putting it into practice. I am pretty sure we are there in five years. I also expect that the awards start to “write themselves” – compiling all the procedural history and summarizing parties’ submissions could be taken over by artificial intelligence. But I do believe that the human brain cannot – and should not – be replaced, not even in 20 years from now. The decider must be a real person. At least with our today’s way of thinking I believe that, even though artificial intelligence can be smarter than anyone, being fair and empathic is attributable to human beings only.

That was all very helpful. Thanks a lot Triinu!