COMMANDTALK

CommandTalk is a system developed by DARPA and the military in the US and is written using a combination of C and Prolog. It's aim is to provide a simulated battlefield and give a commander the ability to direct and control his troops via a speech interface. It uses the Nuance speech recognition tool and a NLP module called Gemini to parse and interpret the utterances, and the ModSAF military battlefield simulator [7,17].

CommandTalk has three separate or distinct grammars; one for controlling an Army, one for an Air-Force and one for a Navy. Each of the grammars is large, complex and allows for a wide range of commands to be issued by the user of the system.

The interface allows a user to create forces and control measures, assign missions to forces, modify missions during execution and control the simulation system functions using ordinary spoken English.

An example sequence of commands issued by a commander is shown in Figure 2.15.

Figure 2.15: Sample CommandTalk dialogue

The Gemini tool that provides the NLP functionality of the application is capable of dealing with the following:

• Noun phrase resolution - conferring with the simulator to keep track of existing units
• Predicate resolution - taking into account difference in carrying out an action depending on the unit involved
• Temporal resolution - using phrasing and context, sometimes explicit indicators can be used
• Vagueness resolution - using linguistics and situational context (and defaults)

An example of the NLP tool resolving issues it cannot work out from context is shown in Figure 2.16. In the example, the system had to realize that it referred to the description of Objective B rather than Objective B itself.

Figure 2.16: CommandTalk dialogue showing resolution