NLP vs. Conversational AI


To date, commercial NLP research has been focused on understanding the many ways people say commands like "Set my alarm for 7 am tomorrow" or "Find a good pizza place nearby." To unlock the power of conversational computing, new technology is required that extracts semantics across multi-turn natural language exchanges, and maintains contextual understanding over time. With conversational AI, people can accomplish transactions that would be impossible with today's command-based virtual agents.

Our AI Components


Semantic Machines is developing a new, language-independent technology platform that goes beyond understanding commands to understanding conversations. Our approach represents a powerful new paradigm, enabling computers to communicate, collaborate, understand goals, and accomplish tasks.

Leveraging extensive proprietary machine learning technology, we are developing the following core components:


Conversation Engine

The Semantic Machines Conversation Engine is a revolutionary approach for modeling human discourse fluidly across speech and text. The engine extracts semantic intent from natural input (voice or text), and then spawns a self-updating learning framework for managing dialog state, context, salience, and end user goals. The Conversation Engine's natural language generation (NLG) technology actively formulates communication with the user based on dialog context.


Deep Learning

Semantic Machines is developing state-of-the-art neural network systems for use across a range of critical functions from semantic analysis to dialog state to acoustic and language models, NLG and speech synthesis.

Speech Recognition

The speech team at Semantic Machines previously led ASR development for Dragon Systems, VoiceSignal, Nuance and Siri at Apple. Now we are building a new speech platform to overcome the limitations of these previous systems. Our ASR technology represents an exciting new advancement providing unique capabilities and performance needed for true conversational computing.

Speech Synthesis

Speech synthesis is critical for conversational computing. The computer’s voice takes the place of a display providing information users need. Existing speech synthesis technology, particularly the prosodic models, are not sufficient to enable effective conversational computing. Leveraging our extensive NLP and machine learning expertise we are developing a proprietary synthesis technology that will enable conversational computing for the first time.

Reinforcement Learning

Reinforcement learning is a core component of our platform. Our novel learning technology enables the system to absorb knowledge from users to continually expand its capabilities in real-time. This learning feedback loop allows the system to improve its understanding of semantics and learn new domains at an ever-increasing rate. 



Semantic Machines is developing the world’s first large-scale training corpus for machine learning spoken and written dialog. The proprietary technology developed to enable the capture, automatic annotation and alignment of data at scale has created an exciting opportunity to make systems that behave the way humans do. Our dataset is key to developing conversational computing models.


Language Independent

The conversational AI technology we are developing is based on a language independent architecture. Our speech and language understanding technology is initially being developed in English, but supports others, including tonal languages like Mandarin.

Yes, 我们的技术支持中文的.


Developer Tools

To enhance and customize the capabilities of our conversational AI we are creating a suite of tools to be used internally and by our partners. Using these tools, developers will be able to adapt our conversational AI technology to their own application space, as well as teach new skills within existing domains.