1. Home
  2. Science

A Game-Changing Approach
To Natural Language Understanding

Proven Approach to NLU

Inspired by the way the brain processes information

Highly Accurate

Understands the meaning of text

Highly Efficient

Requires little training data and processes large amounts of text in real time

Flexible & Transparent

Applicable to any language & easily inspectable

Biologically-Inspired Natural Language Understanding (NLU)

Inspired by neuroscience, Semantic Folding is a machine learning methodology for creating language models with small amounts of training data.

With Semantic Folding, text is converted into a new data representation called a semantic fingerprint. Semantic fingerprints capture the different meanings of words based on thousands of parameters and form clusters of similar contexts. For document processing tasks like classification and semantic search, the system just needs to measure semantic overlaps between semantic fingerprints – a highly efficient computational approach that delivers accurate results with less computing resources.

Semantic Folding is a real differentiator for natural language understanding as it enables a unique combination of high accuracy, efficiency, flexibility and transparency.

What is Semantic Folding?
High Efficiency AI for Text Processing

Meaning-Based Algorithm Delivers High Accuracy

Semantic fingerprints leverage 16k parameters to encapsulate the different meanings of words, sentences or paragraphs. This enables to disambiguate text at a fine-grained level and to match phrases that have similar meanings but different phrasing. As a result, companies get fewer false positives and require less manual intervention to review and correct results.

Semantic Folding Enables Highly Efficient Text Processing

Language models based on semantic fingerprints need orders of magnitude less annotated examples than Transformer models to reach comparable levels of accuracy (a few hundreds versus several thousands). This means that companies get actionable results much quicker, and with less human resources involved.

Semantic fingerprints are sparse distributed vectors that require much less computing power to process large amounts of text. Applications leveraging Semantic Folding do not only save IT costs, they also contribute to lower the carbon footprint of IT operations.

Inspect Results At The Press Of A Button

With Semantic Folding, it is easy to debug and fine-tune language models because each semantic feature can be inspected at the document level. This aspect also guarantees the full transparency and explainability of results, enabling enterprises to eliminate biases in their models and to meet compliance requirements (e.g. GDPR).

Natural Language Understanding That Scales

Semantic Folding empowers business users to customize and train their models with comparatively little example documents. As a result, companies can implement a NLU project where only little training data exist, and easily scale it to other use cases and departments within the enterprise without the need for dedicated, internal AI expertise.

Semantic Folding can be applied to any language and enables direct cross-language text processing, so that translation efforts become obsolete.

Advantages of Semantic Folding

  • High Accuracy

Semantic fingerprints leverage a rich semantic feature set of 16k parameters, enabling a fine-grained disambiguation of words and concepts.

  • High Efficiency

Semantic Folding requires order of magnitude less training material (100s vs, 1’000s) and less compute resources because it uses sparse distributed vectors.

  • High Transparency & Explainability

Each semantic feature can be inspected at the document level so that biases can be eliminated in the models and results explained.

  • High Flexibility & Scalability

Semantic Folding can be applied to any language and use case, and business users can easily customize models.

ChatGPT and LLMs:
The Holy Grail of Enterprise AI?

Read the Article

Explore Semantic Folding

Semantic-Folding-Natural-Language-Understanding

Read the White Paper

Attributions

By Xavier Gigandet et. al. – Gigandet X, Hagmann P, Kurant M, Cammoun L, Meuli R, et al. (2008) Estimating the Confidence Level of White Matter Connections Obtained with MRI Tractography. PLoS ONE 3(12): e4006. doi:10.1371/journal.pone.0004006, CC BY 2.5, https://commons.wikimedia.org/w/index.php?curid=8134159

By Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ, Sporns O – Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ, Sporns O (2008) Mapping the structural core of human cerebral cortex. PLoS Biology Vol. 6, No. 7, e159.[1], CC BY 3.0, https://commons.wikimedia.org/w/index.php?curid=6246097