A System for Answering Simple Questions in Multiple Languages — part 4/4

Cleaned-text code points [43725, 47439). Bibliography omitted here; source.md includes it.

Priyanka Sen, Alham Fikri Aji, and Amir Saffari.
2022. Mintaka: A complex, natural, and multilin-
gual dataset for end-to-end question answering. In
Proceedings of the 29th International Conference
on Computational Linguistics , pages 1604–1619,
Gyeongju, Republic of Korea. International Com-
mittee on Computational Linguistics.
Yiming Tan, Dehai Min, Yu Li, Wenbo Li, Nan Hu,
Yongrui Chen, and Guilin Qi. 2023. Evaluation of
chatgpt as a question answering system for answering
complex questions. CoRR, abs/2303.07992.
Thomas Pellissier Tanon, Denny Vrandecic, Sebastian
Schaffert, Thomas Steiner, and Lydia Pintscher. 2016.
From freebase to wikidata: The great migration. In
Proceedings of the 25th International Conference on
World Wide Web, WWW 2016, Montreal, Canada,
April 11 - 15, 2016, pages 1419–1428. ACM.
Denny Vrandecic and Markus Krötzsch. 2014. Wiki-
data: a free collaborative knowledgebase. Commun.
ACM, 57(10):78–85.
Quan Wang, Zhendong Mao, Bin Wang, and Li Guo.
2017. Knowledge graph embedding: A survey of
approaches and applications. IEEE Trans. Knowl.
Data Eng., 29(12):2724–2743.
Zhiyong Wu, Ben Kao, Tien-Hsuan Wu, Pengcheng Yin,
and Qun Liu. 2020. PERQ: predicting, explaining,
and rectifying failed questions in KB-QA systems.
In WSDM ’20: The Thirteenth ACM International
Conference on Web Search and Data Mining, Hous-
ton, TX, USA, February 3-7, 2020, pages 663–671.
ACM.
Linting Xue, Noah Constant, Adam Roberts, Mihir Kale,
Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and
Colin Raffel. 2021. mT5: A massively multilingual
pre-trained text-to-text transformer. In Proceedings
of the 2021 Conference of the North American Chap-
ter of the Association for Computational Linguistics:
Human Language Technologies, NAACL-HLT 2021,
Online, June 6-11, 2021, pages 483–498. Association
for Computational Linguistics.
Yucheng Zhou, Xiubo Geng, Tao Shen, Wenqiang
Zhang, and Daxin Jiang. 2021. Improving zero-shot
cross-lingual transfer for multilingual question an-
swering over knowledge graph. In Proceedings of
the 2021 Conference of the North American Chap-
ter of the Association for Computational Linguistics:
Human Language Technologies, pages 5822–5834,
Online. Association for Computational Linguistics.
534

A Parameter Settings and Prompts for OpenAI Models
The parameters for GPT-3 (davinci-003) and ChatGPT (GPT-3.5-turbo-0301) were configured to a
temperature setting of 0.1, while the top_p for GPT-3 was set to 0.85. The prompts for both models, which
are illustrated in Figure 3, were used respectively: for GPT-3 directly, and for ChatGPT, conveyed via the
“user” field in the chat API interface.
Figure 3: Prompts employed for ChatGPT and GPT-3 in the English and Russian language benchmarks, respectively.
B Additional Illustrations of the Knowledge Graph Question Answering System
This section contains three illustrations of the graphical user interface with three additional questions as
well as a demonstration of the API for integration into external applications.
Figure 4: Graphical user interface of KGQA system for the one-hop question “What is capital of Germany?”.
535

Figure 5: Graphical user interface of KGQA system for the one-hop question “Who was director of Star Wars?”.
Figure 6: Graphical user interface of KGQA system for the one-hop question “Who is the king of United Kingdom?”.
536

Figure 7: Swagger API for the developed system allowing for integration of the KGQA functionality into applications
(e.g. “seq2seq” or “m3m” endpoints) as well as subcomponents, such as NER for questions or type selection.
Figure 8: Swagger API of an individual endpoint: parameters for KGQA method are documented and can be called
using a RESTful endpoint.
537