IBM's Controlled English modelling language resurfaces on Hacker News

A modelling language written as English

An experimental language that expresses data models in ordinary-looking English has drawn renewed attention after surfacing on Hacker News's front page. The project lives in the ce-store GitHub organisation, which describes itself as an IBM-initiated open source group providing a research environment for the ITA Controlled English (CE) language.

According to the repository's documentation, CE restricts English into a disciplined subset for describing concepts, the properties they carry and how they relate to one another, with one defining constraint: anything written in it should be readable by both humans and machines. The CE Store packages the tooling needed to actually run such an environment.

How sentences become models

The documentation explains that a CE model holds all the information about the data loaded into the store, and that it is built entirely from sentences. Each sentence ends with a full stop and comes in two basic kinds: one defines a concept — a type of thing — while the other defines an instance, a single occurrence of that thing.

The README's worked example first defines a planet concept with mass and radius values:

conceptualise a ~ planet ~ P that has the value M as ~ mass ~ and has the value R as ~ radius ~.

An instance of that concept then reads:

there is a planet named 'Earth' that has '5.972 × 10^24' as mass and has 6371 as radius.

The punctuation is unconventional, but both lines still scan as English, which is the point: the model and its human-facing description are the same artefact rather than two things to keep in sync.

Ways to run it

The repository lists several deployment routes. A one-click button deploys a cloud-hosted instance to IBM Cloud, which offers free sign-up; the docs note that after deployment the store sits under /ce-store/ on the assigned URL rather than at the root, with the engineering panel reachable at /ce-store/ui/.

For local use, the code can be cloned from GitHub and started through Apache Maven, which runs the application on an embedded Tomcat server at localhost:8080/ce-store. A prebuilt Docker image is available via docker pull cestore/ce-store, and the README also covers building a WAR file with Maven, turning it into a custom Docker image, and a Vagrant-based setup.

The Engineering Panel

Day-to-day work happens in the Engineering Panel, a web interface with four areas: a left-hand pane for loading and modifying CE models, a right-hand object pane for browsing what the store contains, a bottom messages area where errors are reported, and a central working area for interacting with the model. Through the panel, users can load new sentences and run queries against the stored data.

The store ships with a sample set of CE sentences that build a model around medical data, which the project wiki covers in a tutorial, alongside a video walkthrough of the language running just under 23 minutes. The code is released under the Apache License, Version 2.0.

Why it matters

Controlled natural languages occupy an awkward niche: they must be strict enough for software to parse yet natural enough that people can read them without training. CE is one of the more readable attempts at that balance, and its reappearance on Hacker News points to a lasting appetite for knowledge-representation formats that can be inspected without learning a graph query language or an ontology editor.

The dual-readability property is the practical draw. A model written in CE can be reviewed by domain experts who never learn a formal syntax while remaining loadable and queryable by the store's tooling, collapsing the usual gap between documentation and data. The project is openly experimental in intent, so the CE Store is best read as a reference implementation of an idea rather than a finished product — and the renewed discussion around it seems as much about the approach as about the code itself.