Synthesis and Evaluation of a Domain-specific Large Data Set for Dungeons & Dragons

Synthesis and Evaluation of a Domain-specific Large Data Set for Dungeons & Dragons Akila Peiris and Nisansa de Silva Department of Computer Science & Engineering, University of Moratuwa, Sri Lanka {akila.21,nisansadds}@cse.mrt.ac.lk Abstract This paper introduces the Forgotten Realms Wiki (FRW) data set and domain specific natu- ral language generation using FRW along with related analyses. Forgotten Realms is the de- facto default setting of the popular open ended tabletop fantasy role playing game, Dungeons & Dragons. The data set was extracted from the Forgotten Realms Fandom wiki consisting of more than over 45,200 articles. The FRW data set is constituted of 11 sub-data sets in a number of formats: raw plain text, plain text annotated by article title, directed link graphs, wiki info-boxes annotated by the wiki article title, Poincaré embedding of first link graph, multiple Word2Vec and Doc2Vec models of the corpus. This is the first data set of this size for the Dungeons & Dragons domain. We then present a pairwise similarity comparison bench- mark which utilizes similarity measures. In ad- dition, we perform D&D domain specific nat- ural language generation using the corpus and evaluate the named entity classification with respect to the lore of Forgotten Realms. 1 Introduction Specialized and domain specific data sets are useful for a number of advanced tasks in the domain of Natural Language Processing (NLP). For example, recent studies have shown that the domain speci- ficity significantly impacts vital NLP tasks such as measuring semantic similarity (Sugathadasa et al., 2017) and domain specific text generation (Lebret et al., 2016). Further, it has been shown that mod- els developed using data from a generic domain do not seamlessly transfer to tasks in a specific do- main (Rajapaksha et al., 2020). Fantasy domains could be considered an extreme case of domain specific data, as it is possible to observe the full spectrum of deviations from the non-domain spe- cific (general domain) usage, both in the lexical and semantic perspectives. An example for lexical differences is the usage of dwarves as the plural form of dwarf in the fantasy genre1 in place of the general domain spelling dwarfs. An example for semantic differences can be seen in the words Ghost2 and Wraith3. In the Merriam-Webster dic- tionary, they are given as synonyms in the generic domain4 while in the domain of the fantasy role playing game Dungeons & Dragons, they are de- fined as two distinct creatures. In this paper, we present the FRW data set, specific to the Forgotten Realms setting of Dungeons & Dragons. We expect our data set to be useful for in-domain tasks such as text generation (Zhang et al., 2019), information extraction (de Silva and Dou, 2021), and informa- tion retrieval (Sugathadasa et al., 2018). We also anticipate our data set being vital for cross domain tasks such as text alignment (Sanchez-Perez et al., 2014), style transfer (Fu et al., 2018), and summa- rizing (El-Kassas et al., 2020). As a primer for these usages, we introduce a pairwise similarity …

Content truncated to 500 words