The best teachers are the ones who can understand and adapt to their students’ needs and abilities. Cyc’s Conversational Training Tool employs best-in-breed NLP and NLU capabilities alongside our top-notch user modelling framework to help students get the most of their training.
Understanding a student’s needs requires a robust understanding of what information they need to know, how they intend to use the information, and how they will build on it in the future. A surgical scrub nurse and a surgeon may need to know about the same equipment, but there will be pieces of knowledge that are required for one career but not the other. Even information that is essential to both should be presented differently to trainees for the different roles so that it can be appropriately contextualized and integrated with the rest of their training. Cyc is able to represent all of this information in its internal language and reason about both what should be presented to the trainee and how it should be presented.
It is similarly crucial to understand how much the trainee currently knows. Cyc is able to gauge this using a variety of sources. Cyc can track the trainee’s progress through the training application by reasoning about what modules have been completed, which questions the trainee has asked the system, and how they have performed on training questions. Because Cyc can easily integrate information from external data sources, Cyc can also incorporate information about a student’s performance on external training exercises and assessments.
Evaluation of a student’s performance is not limited to whether they answered a question correctly. Cyc can also reason about how they may have narrowed the answers or, in the case of incorrect answers, why they may have answered as they did. When asked which medication should be administered to a hypothetical patient, a nursing student who first correctly narrows the answers to ACE inhibitors before selecting the wrong one likely knows more than a student who guesses a different sort of medication entirely.
Cyc is able to combine all of this information to provide a conversational training experience. At each conversational turn, it reasons about whether it should quiz the trainee on recent concepts, review with them their past performance, or offer further clarifying information. Whichever option it chooses, Cyc always engages with the trainee in a way that is appropriate to their current understanding and training goals.
Cyc’s commonsense-driven NLU capabilities make these conversations feel natural and effortless. Cyc is able to understand each user utterance as situated in a context, allowing it to reason through possible ambiguities and resolve descriptive and anaphoric reference. In a context where Cyc and the trainee are discussing ibuprofen, they need not continually refer to it as “ibuprofen”, but could more naturally say “the NSAID”, “it”, or, “the medication”.
This fits in with Cyc’s broader effort towards Knowledge Base Natural Language Understanding (KBNLU). Any adequate solution to the problem of NLU needs common sense. Consider:
The trophy didn’t fit into the suitcase because it was too large.
What is the referent of “it”? Now consider:
The trophy didn’t fit into the suitcase because it was too small.
Notice that swapping out “large” and “small” changes the referent of “it” from the trophy to the suitcase. This is not due to some syntactic trick. Rather, your interpretation of the sentence is informed by your knowledge of the world. Cyc’s KB provides that knowledge to power the KBNLU effort.
This technology shines in a training context. Cyc is able to interact with users in ways that are intuitive. And, while Cycorp hasn’t solved the NLU problem yet, we are meeting our milestones on that journey. Meanwhile, every step forward has direct, positive impacts on all of Cyc’s applications, particularly those requiring conversational fluency.