Update README.md

README.md

@@ -1,6 +1,62 @@
+# Flexudy's Conceptor: Towards Neuro-Symbolic Language Understanding
+
+![alt text](https://www.flexudy.com/wp-content/uploads/2021/09/conceptor.png "Flexudy's conceptor")
+
+At [Flexudy](https://flexudy.com), we look for ways to unify symbolic and sub-symbolic methods to improve model interpretation and inference.
+
+## Problem
+
+1. Word embeddings are awesome 🚀. However, no one really knows what an array of 768 numbers means?
+2. Text/Token classification is also awesome ❤️‍. Still, classifying things into a finite set of concepts is rather limited.
+3. Last but not least, how do I know that the word *cat* is a **mammal** and also an **animal** if my neural network is only trained to predict whether something is an animal or not?
+
+## Solution
+
+1. It would be cool if my neural network would just know that **cat** is an **animal** right? *∀x.Cat(x) ⇒ Animal(x)*.
+Or for example, (*∀x.SchöneBlumen(x) ⇒ Blumen(x)*) -- English meaning: For all x, If x is a beautiful flower, then x is still a flower. --
+
+2. All of a sudden, tasks like **Question Answering**, **Summarization**, **Named Entity Recognition** or even **Intent Classification** etc become easier right?
+
+Well, one might probably still need time to build a good and robust solution that is not as large as **GPT3**.
+
+Like [Peter Gärdenfors, author of conceptual spaces](https://www.goodreads.com/book/show/1877443.Conceptual_Spaces), we are trying to find ways to navigate between the symbolic and the sub-symbolic by thinking in concepts.
+
+Should such a solution exist, one could easily leverage true logical reasoning engines on natural language.
+
+How awesome would that be? 💡
+
+## Flexudy's Conceptor
+
+1. We developed a poor man's implementation of the ideal solution described above.
+2. Though it is a poor man's model, **it is still a useful one** 🤗.
+
+### Usage
+
+No library should anyone suffer. Especially not if it is built on top of **HF Transformers**.
+
+
+Go to the [Github repo](https://github.com/flexudy/natural-language-logic)
+```python
+from core.conceptor.start import FlexudyConceptInferenceMachineFactory
+
+# Load me only once
+concept_inference_machine = FlexudyConceptInferenceMachineFactory.get_concept_inference_machine()
+
+# A list of terms.
+terms = ["cat", "dog", "economics and sociology", "public company"]
+
+# If you don't pass the language, a language detector will attempt to predict it for you
+# If any error occurs, the language defaults to English.
+language = "en"
+
+# Predict concepts
+# You can also pass the batch_size=2 and the beam_size=4
+concepts = concept_inference_machine.infer_concepts(terms, language=language)
+```
+
 Output:
 
-``{<br/>'cat': ['mammal', 'animal'], <br/> 'dog': ['hound', 'animal'], <br/>'economics and sociology': ['both fields of study'], <br/>'public company': ['company']<br/>}
+```{<br/>'cat': ['mammal', 'animal'], <br/> 'dog': ['hound', 'animal'], <br/>'economics and sociology': ['both fields of study'], <br/>'public company': ['company']<br/>}```
 
 ### How was it trained?``