Automated Generation of Sports Reports – Interview with Harald Salg
Medien-Service Untermain GmbH (MSU) is part of a strong, traditional media group with headquarters in Aschaffenburg and has been creating state-of-the-art software solutions for media houses since 1996. In addition, they provide their customers with comprehensive information on almost all team sports represented in Germany within the framework of the MSUsport result reports. In order to deliver not only tables, but also complete sentences and texts, the company uses the software of AX Semantics.
What do you use the automated content generation software for?
The goal we pursue with the automated content generation is to create lively sports reports from a huge treasure of sports-related data. We want to encourage our customers to move away from the purely numbers and statistics-oriented output of, for example, soccer or handball matches, leagues or competitions to the narrative reproduction of this content.
We firmly believe that such content brings much more value, especially for media houses. Of course, it is not possible for our customers in the traditional environment to produce sports reports themselves or to acquire them from freelancers for even a fraction of the leagues issued as tables. We close this gap with our service.
How many texts have you already generated/are you planning to generate?
We started with a first research project in spring 2020 and are now working on perfecting the content generation. During this time, we have already generated several thousand sports reports. Unfortunately, renewed restrictions or bans on sporting events slowed us down here in autumn 2020. Since our goal is not to text professional events but local sports, our hands are currently tied here. We very much hope that we will be able to start again in spring 2021. If our customers accept the service well, it can happen very quickly.
An average customer manages about 200 soccer and 100 handball leagues per season with our system, among many other sports. If you calculate with an average of 10 games per league and match day, that would be about 3,000 sports reports per weekend. With currently about 100 customers, that could become a lot relatively quickly.
What problem/challenge do you solve with the AX Semantics software?
There is no single challenge that we solve with the AX Semantics software. It is a rather unmanageable variety of challenges. What is important here, however, is the creation of a large variance – especially with regard to a large number of texts – and the automation of grammatical constructs, especially with regard to singular/plural and gender forms.
What plans do you have for the future (more text types, languages, etc.)?
Currently we have trained the AX software for the sports handball and soccer and especially for match reporting. We will continue to develop these variants and implement greater variance, different text lengths and new content types. To this end, we will also expand MSUsport in terms of statistics generation and automated match analysis.
Apart from that, other sports – I personally would be fascinated by basketball, for example – are on the agenda for us first. In parallel, we plan to move into pre-match reporting. This means that our software will analyse the upcoming matches of the next match day and we would like to write articles on this basis.
How would you describe the AX Semantics software in one sentence to someone who doesn't know the tool yet?
I don’t know if I would describe your software or rather your self-service approach and especially the excellent customer support. I have been in software development for over 25 years and have seen many excellent products die, because the best software is not worth much without the appropriate support. With your tool and tutorials it was absolutely easy to get started and then at that point the support made the seamless transition. From my point of view, it couldn’t have worked any better.
About AX Semantics
AX Semantics is a user-friendly Natural Language Generation (NLG) software that makes content generation scalable – in 110 languages. Unique and meaningful sports reports can be generated based on data. The software is 100 percent SaaS – meaning that all functions are available via the application in the web browser. No programming knowledge is required to use it.