Example project: requirements analysis
Let's take a look at some of the considerations that we took into account when we created the Performance Analytics accelerator.
In creating this project, we considered the following:
Consulting SMEs. The subject-matter experts we talked to included our internal pre-sales team who were able to give us input that came directly from Arria clients. For example, the SMEs are the ones who determined which sectors we mention as sales drivers and offsets.
Understanding edge cases. What do we say if overall growth was negative or flat? How does this impact the wording of the narrative? What if one of the individual sectors mentioned under offsets had negative or flat growth? Would anything change in the narrative? Developers need to understand how unusual situations are presented in a narrative.
How to handle errors. What should the NLG Studio project do if there are problems in the data? For example, some cells in the spreadsheet are empty. Again, developers need to understand how to deal with these.
How to ensure consistency. A key challenge in requirements and corpus analysis is that developers get inconsistent examples and advice from different people. We could illustrate this by showing an alternate narrative, such as
Actual Sales went up by $104.83M — a 247.93% increase from 2015 to 2016. The best performers were USA ($23.66M), Ireland (23.42M), and Belgium ($22.09M).
Here we have differences in wording and in information presented. Developers need to look at various options for narratives and decide which to use (or how to combine them), and indeed where it makes sense to include linguistic variation.
