Voice-Driven Modeling: Software Modeling Using Automated Speech Recognition

AI-driven streamlined modeling: experiences and lessons learned from multiple domains

“…We classify and review AI-driven model specification approaches by which artifacts are available to conduct the specification and activities involved in the specification (including domain-specific languages and domain-specific modeling languages) rather than specific formats or formalisms arrived at during the activity. Accordingly, we classify AI-driven approaches for model specification as -(a) voice-driven specification approaches [17,47] and (b) collaborative specification approaches using social networks and chatbots [55]. We review these next.…”

“…Voice-driven modeling proposes to use speech processing and NLP to achieve context-specific modeling [17]. Primarily motivated to increase the efficiency of modelers and to enable modeling by persons with disabilities [47], it builds on similar efforts in voice-driven or vocal/spoken programming [86].…”

“…Any legal case is likely to have specific instances of these concepts [80]. For cases such as parental alienation (a major concern in Western Europe), we obtain the geography-specific open case data 17 . We apply the CMG (discussed in Sect.…”

“…Applicability and Customer Buy-in For the actor-based enterprise simulation with reinforcement learning, customer interactions and buy-ins involve a supermarket chain, a large postal company, and a telecommunications company in Western Europe. RW requirement documents/ problem description documents [7,65], repositories of (Meta-) models/ ontologies [5,69], external knowledge sources [2], dependency graphs [7], Tagged text [65], relational and graph databases [2] chat/text messages [55], voice commands [17,47], textual and visual DSL scans [56] repositories of (Meta-) models [23,44] sensor data [37], process data [59], clinical data [39], multi-lingual bio-medical data [31] graph databases [37,57] CS regulatory documents [75,77], standards and generated documents [54], legal case texts [80] rule sentences [63] SBVR models [75] historical formulations, Wikipedia + specialty sites [66,81] AI Techniques RW lexico-syntactic pattern matching [2,7], co-occurrence analysis [2], syntactic parsing and rule-based processing [7] classification with text vectors [65] speech recognition and synthesis [47], NLP/NLU [17] classification of (meta-) models using deep NN [23] semantic annotations+ cross-lingual concept matching [31], finite state machine over ontology [39] CS clustering…”

“…Model specification and transformation It can be observed in Table 1 that the nature of artifacts used and created as well as the applicable techniques in these activities differ substantially from the domain modeling activity. There is a natural shift towards alternate ways of model specification such as chat messages [55] and voice commands [17,47] and techniques such as speech recognition [17], natural language understanding and collaborative specification [55], optical character and image recognition [56], etc.…”