Arria is a proud sponsor of the annual International Conference on Natural Language Generation (INLG), which this year took place 20-24 September 2021. INLG 2021 was hosted by the University of Aberdeen, and while most presentations were delivered virtually this year, there were some in-person activities as well. With the gathering’s participants being mostly NLG researchers, it’s no surprise there was a lot of exciting discussion and presentation of new ideas for NLG.
Arria staff contributed to INLG in many ways:
Ross Turner (Arria’s Chief Product Officer) was on a panel about “What users want from real-world NLG”. Veering away from the research perspective, Ross and the other panelists tried to give a pragmatic view on what commercial users of NLG really want (which is not always what researchers focus on). The panel discussed trust — how users need to trust NLG systems — and hence emphasized the importance of quality assurance and transparency. Users also need the ability to configure and control NLG systems. More generally, people who do not have PhDs in AI need the ability to build NLG systems. The panel was followed by a lively discussion, and was an eye-opener for many attendees, including PhD students who had little previous exposure to the “real-world” NLG perspective.
Ehud Reiter (Arria’s Chief Scientist) was a co-author on several papers, including a shared task that encouraged researchers to learn new approaches to finding mistakes in NLG texts, a shared task on reproducibility of human evaluations, and a paper on using NLG to explain machine learning. The work on new approaches to finding mistakes was especially well received, and we hope that this will encourage the research community to think more about innovative approaches to quality assurance in NLG.
Stephanie Inglis (a member of Arria’s Solutions team) was a co-author on a paper on error analysis for NLG, which won an award as one of the top papers at INLG. Steph and her colleagues gave recommendations on how NLG researchers and developers could analyze mistakes in generated narratives in order to understand weaknesses in NLG systems and hence how these could be improved.
Arria’s Aberdeen Office hosted a coffee morning on 23 September, to which all INLG attendees in Aberdeen were invited. Ehud Reiter and Neil Burnett (Chief Technology Officer) gave an introduction to Arria, and attendees then had a tour of Arria’s new office (which is in a beautiful old church with stained-glass windows). The coffee morning also featured demos and a chance to chat to Arria staff. A well-attended event, the get-together was enjoyed by both Arria’s staff and our visitors.
Overall, INLG 2021 was a great success, and we look forward to INLG 2022!
For more information, Ehud Reiter has provided a thorough summary of the discussion panel, on his own blog, in INLG: What real-world NLG users want.