A new document arrives.
What does it change?
Find the relevant passage, compare accounts and examine the decisions and interactions behind a situation.
An email arrives after you have already spent hours reconstructing a situation. It mentions a meeting, a postponed decision or information passed to someone else. You remember reading something similar. In which document? Was it the same event? And what does it change in your understanding of the case?
This difficulty guides a specific area of work at OpenProof: helping you find useful passages from the situation you are trying to understand, including when you no longer know a document’s name or where it was filed.
01
One event can appear in several documents
Consider an entirely fictional example from a company.
- DOCUMENT 01
Minutes written on 12 March
“At the meeting on 8 March, approval was postponed.”
Distinguish the date of the minutes from the date of the meeting.
- DOCUMENT 02
Email sent on 9 March
“At yesterday’s meeting, we asked for the figures before making a decision.”
Examine whether the email concerns the same meeting and what it adds.
- DOCUMENT 03
Note written on 20 March
“We finally approved the budget at the second meeting.”
Keep the second meeting as a separate event.
Documents 01 and 02: a connection to check. Document 03: a different step.
The first two documents may help explain the same event. The third describes a later step. A useful connection needs to preserve that distinction and show the reasons for linking the passages.
02
From a passage to a more precise question
Who could approve the budget? Who had the figures? When were they shared? Was approval postponed because information was missing, because of a procedural constraint, or for another reason?
These questions guide the examination of the documents. Similar wording or the presence of the same people does not establish a shared intention. An explanation needs to be considered alongside supporting material, contradictory material and what remains missing.
This is where document research connects with an examination of the actors’ decisions and interactions: the information they had, the constraints they faced and the effects of their choices. Several explanations may need to remain open when the sources cannot settle the question.
03
One original, several passages and several interpretations
Our design principle is to preserve the original document and reference precise passages from it. The same document may inform an event in a chronology, a question about an actor and an observation in the analysis.
A person taking over the case should be able to return to the original passage and understand why it was selected. A correction should remain identifiable, together with the interpretation it changes. A statement repeated in several documents should also retain its visible origin.
04
What we are testing
We are working towards a workflow in which a question can bring together passages scattered across documents, with an explicit reason for each proposed connection. Users can accept a connection, reject it or leave it open.
Our evaluation needs to check whether the system finds the relevant passages, keeps similar events separate, surfaces a contradictory document and preserves corrections after reopening and export.
The intended benefit is practical: helping you understand what a document adds, what remains uncertain and which check could move the case forward. For a lawyer taking over the matter, this means being able to examine the sources and the reasoning connecting them.
05
ECB+ and TimeML: two references informing our work
Agata Cybulska and Piek Vossen’s work on the ECB+ corpus examines event coreference: deciding whether passages refer to the same event when similar events are also present. For someone reviewing a case, the question is practical: do these two descriptions concern the same meeting?
TimeML, described by James Pustejovsky and his co-authors, distinguishes events, time expressions and the relations connecting them. It can also represent partially specified time references. This informs our requirement to preserve uncertainty rather than assign artificial precision.
We use these questions to inform OpenProof’s design and evaluation. Results obtained on news articles do not establish our reliability on French case documents. These public references do not constitute validation of the software or collaboration with their authors.
YOUR NEXT STEP
Discover OpenProof. 8 days free.
No card required, no payment to start and no automatic subscription.
Try OpenProof — 8 days free ↗The trial currently runs in a test environment: use fictional documents only.
Already have an account? Sign in ↗
