How Ask Lutecium Works
Ask Lutecium is a semantic search system. Unlike a conventional search engine, it does not look only for the exact words entered by the user. It attempts to find passages whose meaning is close to the meaning of the question.
From documents to passages
The Lutecium corpus contains documents of very different sizes: WordPress pages and posts, citations from the Mathèmes de Lacan, and documents preserved in selected Lutecium mirror archives.
For semantic search, long documents are divided into smaller units called passages. A passage contains at most about 220 words. Consecutive passages overlap by about 40 words. This overlap helps preserve sentences or ideas that cross the boundary between two passages.
For example, a 500-word document is divided approximately as follows:
- Passage 1: words 1–220
- Passage 2: words 181–400
- Passage 3: words 361–500
Each passage remains linked to its original document and source.
A 384-dimensional semantic space
Each passage is analysed by a multilingual text-embedding model called E5. It is a model made with a Transformer Neural Network.
The model transforms a passage into a list of 384 numbers. Mathematically, these numbers can be regarded as the coordinates of a point in a 384-dimensional space. Presently it uses 1.5 millions passages or sequences.
This is analogous to locating a point in ordinary two-dimensional space with coordinates (x, y), or in three-dimensional space with (x, y, z), except that here there are 384 coordinates (x1, x2, x3, x4, …, x384).
The individual dimensions do not correspond to predefined concepts such as “Lacan”, “desire”, “topology” or “Freud”. Their meaning has been learned by the model during training and is distributed across many dimensions simultaneously. What matters is therefore not one particular dimension, but the relative position of complete vectors.
How a question is searched
When a visitor enters a question, Ask Lutecium transforms the question into another 384-dimensional vector using the same E5 model.
It then compares this vector with the vectors representing the Lutecium passages. Passages whose vectors point in similar directions are considered semantically close and are ranked near the top of the results.
This means that the words used in the question do not necessarily have to occur literally in the document. For example, a question concerning “the constitution of the ego through the specular image” may retrieve passages concerning “the mirror stage”, even if the exact words of the question are absent.
Multilingual search
The E5 model used by Ask Lutecium is multilingual. A question written in English may therefore retrieve a relevant passage written in French, and a French question may retrieve relevant English material.
Sources remain visible
Semantic similarity is not itself a scholarly interpretation or a statement of authority. 為此, Ask Lutecium preserves the provenance of every result.
The semantic corpus includes several distinct families of material:
- Lutecium WordPress pages and posts
- the Mathèmes de Lacan
- selected documents preserved in Lutecium mirror archives
These sources are not merged into an anonymous mass: their origin remains identifiable, and results link back to their original Lutecium source whenever possible.
Semantic search and exact search are complementary
Ask Lutecium does not replace the other search tools available on the site.
The ordinary Lutecium search remains useful when looking for a particular author, title, expression or exact word.
The specialized Mathèmes search remains preferable when searching exact Lacanian terminology, symbolic notation, dates, bibliography references, indexed terms, or combinations of precise terms.
Ask Lutecium is most useful when:
- the question is conceptual;
- the exact vocabulary used in the source is unknown;
- related documents may use different formulations;
- relevant material may be written in another language.
In short, conventional search asks: “Where do these words occur?”
Ask Lutecium asks something closer to: “Which passages in Lutecium are nearest in meaning to this question?”