Leader in symbolic, and deep learning language processing

A natural language compiler is the process of translating a natural language input into a formal language. This formal statement could be a SQL query, a JSON object, or an API call. This complex problem is best looked at with a combination of formal descriptions for which we use the CLOVER programming language and generative LLMs.

It is possible to reduce the size of LLM, and hence its cost, by one to two orders of magnitude (depending of the type of LLM) by injecting high quality pre-analysis to training and runtime input. This is a process we call hybridization. It leverages the incredible predictive power of LLMs with the ability to interpret part of the input with some symbolic certainty.

Traditional NLP tasks, such as Named Entity Recognition (NER), focus on the extraction of sub-strings, letting the mapping of those substrings to actual identifiable entities to other processes. We believe these two steps need to be combined to become part of the same feedback loop. This leads to actionable output, including actual identifiable persons, objects, and predicate statements on those identifiable entities. Such actions include API calls.