A few weeks ago Narrative Science announced that it would be bought by Salesforce and become part of the Tableau team. Tableau is a business intelligence and data visualisation tool which was bought by Salesforce in 2019. To my knowledge, this is the first time that an NLG company has been bought by one of the major IT/tech companies (Salesforce isnt as big as Google, but its still huge!).
Anyways, I thought I’d write a few thoughts about this. I should say that I have no special knowledge of the above deal, my knowledge is purely based on press releases and public blogs.
I think most people would consider Narrative Science (as well as Arria) to be one of the leading companies which specialise in NLG. My understanding is that they started as a spinoff from the journalism school at Northwestern University, with their academic roots being in computational creativity instead of NLP/NLG. They initially focused on journalism and sports reporting applications, but over time migrated to focusing on financial reporting and business intelligence.
From a personal perspective, one odd thing is that I have never met or interacted with anyone from Narrative Science. The commercial NLG world is a small one, and over the past ten years I have interacted with people from other NLG companies such as Ax Semantics and Yseop. I’ve met them at academic conferences such as INLG (including a very interesting industry panel this year) and professional events, welcomed their occasional comments on my blogs and papers, and commiserated over shared gripes like crazy client expectations caused by GPT3 hype. So its a bit strange that I’ve never interacted with anyone from Narrative Science.
And probably for similar reasons, I know very little about Narrative Science’s NLG technology or approach. I know a bit about what Ax Semantics and Yseop are doing; not because I have access to anything secret, but because I know (partially through interacting with people) where to look on the public web to find out about their products. Since I dont know anyone at Narrative Science, I’m left with looking at the marketing material on their website, which doesnt tell me a lot about what they are actually doing.
Anyways, although Narrative Science remains something of a mystery from a people and technology perspective, they have done a lot to “spread the word” about NLG and what it can do to the wider business and commercial world, and I am grateful to them for this.
I think the most interesting aspect of this development is that Narrative Science is being bought by Tableau, which sells business intelligence tools. Last year, Microsoft, which is also a leading vendor of business intelligence tools, announced that it was adding a “SmartNarrative” feature to its PowerBI tool. The technology forecasting firm Gartner has been saying for a while that NLG will become a key feature in business intelligence, maybe this is driving the above actions?
People sometimes say that NLG needs a “killer app”. I think the business world is telling us that business intelligence could be such a killer app! For people who are not familiar with NLG for business intelligence, its essentially about generating texts that communicate important insights about business data. A simple example is the Tibco-Arria Covid dashboard. There are plenty of other examples on the Arria and Narrative Science websites.
I’d love to see more research in the academic NLG community on topics that are important in NLG-for-BI, including
- accuracy: NLG texts for business intelligence MUST be 100% accurate and cannot be misleading.
- content selection: it is essential that NLG-for-BI systems be able to select the most important insights to communicate in the generated text.
- articulate analytics: we want analytics which are easily expressed in language; which insights should be verbalised rather than visualised?
- configurability: users of BI visualisation tools expect to have fine control over the details of a visualisation; users of BI NLG tools likewise expect to good control over the details of NLG texts.
- interactivity: users expect BI tools to be highly interactive; this applies to NLG texts as well as visuals.
Of course research should not be completely driven by applications! But I think having real applications and users does provide useful scientific insights, and also makes it easier to get resources. I also think the above questions are very interesting scientifically as well as practically.
I have deliberately not said anything about how this news affects Arria; I cannot reveal commercially sensitive or confidential information about Arria in this blog, and the easiest way to do this here is to avoid the topic. But anyways, I think the key “take home” message for my readers from Tableau’s acquisition of Narrative Science is not the impact on Arria, but rather that this highlights that NLG is becoming important in business intelligence, and this could be a “killer app” for NLG.