Impact of Varying Vocabularies on Controlling Motion of a Virtual Actor

Förger, Klaus; Honkela, Timo; Takala, Tapio

doi:10.1007/978-3-642-40415-3_21

Cited by 2 publications

(2 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This work is inspired by [Ma 2006], which provides an action to a virtual agent from a single sentence. However, our method works for several agents over longer periods of time.Natural Language commands to virtual agents are inherently ambiguous and require several modifications in order to properly command virtual actors [Förger et al 2013]. We constrain the problem domain to remove ambiguities, namely, by not allowing pronouns and requiring all text to be in the present tense.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Agent script generation using descriptive text documents

Balint

Gingold

Allbeck

2014

Proceedings of the Seventh International Conference on Motion in Games

View full text Add to dashboard Cite

When designing games, artists exert large efforts to create visually compelling scenes. Work such as WordsEye [Coyne and Sproat 2001] can assist artists by parsing natural language texts into static scenes. Complementary to this endeavour is the population of that environment by simulation authors. Adding agents to an environment with plausible behaviors is a time consuming process, as most require individual scripts to control their behavior. This generally degrades variability, as scripts are re-used. In order to assist in creating commands and scenes for virtual actors, we propose a method that can create scripts for agents to plausibly act within a virtual environment. This work is inspired by [Ma 2006], which provides an action to a virtual agent from a single sentence. However, our method works for several agents over longer periods of time.Natural Language commands to virtual agents are inherently ambiguous and require several modifications in order to properly command virtual actors [Förger et al. 2013]. We constrain the problem domain to remove ambiguities, namely, by not allowing pronouns and requiring all text to be in the present tense. The first allows our system to easily parse out actors and objects in the scene, and keeps users from writing ambiguous sentences by mistake, hopefully making it easier for a novice user to debug scripts. The second constraint comes naturally from the problem domain, where virtual agents are performing actions to display to a user. We also define three types of actions that an agent can perform: postural, event, and gestural actions. Postural actions control the end state of an agent, event actions update the state of the world, and gestural actions are atomic actions with no effect on the world state.One challenge of using natural language to generate agent scripts is determining when an agent should start performing their action. [Porteous et al. 2010] used decomposition planning to determine approximately when and where an actor should be. However, by using a representation such as PAR [Bindiganavale et al. 2000] we can create a loose ordering, where agents know to perform actions based on the previous agents' actions. When describing a scene, it is common to use temporal phrasing such as Then this happened. We use this information for turn taking. In the absence of temporal phrasing, we assume multiple agents perform an action concurrently with the last action encountered in the script.Actions within a story can also be in progress when the scene begins. For example, Anna is holding the cup implies that the character Anna is performing a holding animation when the user enters into the story. Other actions more obviously occur during a scene, such as George runs to the bathroom. To distinguish these two for rendering purposes, we examine the tense of the verb. In the two aforementioned sentences, the first is a continuous tense and the second is simple present. Continuous actions are realized in the same manner as active verbs, but simply start their realization bef...

show abstract

mentioning

confidence: 99%

“…Natural Language commands to virtual agents are inherently ambiguous and require several modifications in order to properly command virtual actors [Förger et al 2013]. We constrain the problem domain to remove ambiguities, namely, by not allowing pronouns and requiring all text to be in the present tense.…”

mentioning

confidence: 99%