Textual Case Based Reasoning¶
A case-based method that uses text-bearing past cases to retrieve relevant experience or reuse it for a new problem.
Core Idea¶
Textual case based reasoning (TCBR) uses earlier cases recorded in or drawn from natural-language text to help with a new problem. A system relates the current problem to those earlier cases and makes their textual knowledge useful. It may return relevant cases for a person to examine, analyze a new case, or reuse prior solution text in a proposed answer. The text-to-case step is essential; one particular algorithm or a complete four-stage CBR cycle is not.[ref-4ce3bc7a5903][ref-4eebad34b0e0][^ref-894d3703de86]
The legal and email examples below both use past textual experience, but they solve different problems. Their shared identity is the relation among a new task, text-bearing prior cases, a way to make that text usable, and a task-relevant result.[ref-4eebad34b0e0][ref-894d3703de86]
Use Within the Domain¶
To identify TCBR, ask four questions: What new case or request is being addressed? What earlier case knowledge is in text? How is that text related to the new problem? What useful cases, analysis, or proposed response results? A generic text search lacks the case-solving relation; a purely structured case system lacks the textual part.[ref-4ce3bc7a5903][ref-4eebad34b0e0][^ref-894d3703de86]
The system need not do every later step itself. An earlier case retrieved for a defined problem can be useful even if a human carries out the reasoning. In another implementation, software may extract features and generate an analysis or mark text for adaptation. Retrieval, revision, and retention are possibilities, not mandatory stages in every observed instance.[ref-4ce3bc7a5903][ref-4eebad34b0e0][^ref-894d3703de86]
Clarity¶
The name makes a useful boundary explicit: text supplies or shapes case knowledge, and case-based reasoning gives that text a role in a new problem. In SMILE+IBP, legal fact sentences become Factor indices used to compare cases. In the email system, stored requests and replies are retrieved and earlier reply passages are annotated for possible reuse. These are two different bridges from text to cases.[ref-4eebad34b0e0][ref-894d3703de86]
Manages Complexity¶
Earlier cases contain details that may not fit a fixed rule. A text-to-case method can surface or reuse that experience, but it also has limits. Automatic feature extraction may miss case information; a past email reply may contain passages that do not fit the new request. A retrieved case or annotated draft therefore remains material for the task, not a guaranteed final answer.[ref-4eebad34b0e0][ref-894d3703de86]
Abstract Reasoning¶
The reusable pattern is: define the current case, find relevant text-bearing past cases, connect their text to the task, and use the result at the appropriate level. That result might stop at retrieved cases, or proceed to analysis or a proposed solution. This pattern distinguishes TCBR from topical document retrieval while allowing legal prediction and assisted email writing to be examples of the same method.[ref-4ce3bc7a5903][ref-4eebad34b0e0][^ref-894d3703de86]
Knowledge Transfer¶
A new domain can borrow the question “How can earlier textual cases help with this case?” It must then decide what a case is, what parts are textual, how relevance is judged, and who validates the output. Legal Factors are not a ready-made email representation, and email passage annotations are not a legal prediction method. The common method travels only with a new task-specific text-to-case bridge.[ref-4eebad34b0e0][ref-894d3703de86]
Example¶
In a legal study, SMILE accepts text describing a new trade-secret case and converts sentences into Factor indices. IBP compares earlier cases using those factors and produces issue analysis and a prediction or abstention. The new legal summary is the problem; past case texts supply knowledge; factor indexing is the bridge; analysis and prediction are the output. A 146-case evaluation exposed a gap between automatic and manual representation and did not establish deployment.[^ref-4eebad34b0e0]
In an email study, stored request/response pairs are textual cases. For a new request, the system retrieves a past response and marks passages that could be kept, omitted, or changed. The incoming email is the problem; stored messages are cases; retrieval and passage annotation are the bridge; an annotated proposed reply is the output. The human writer has final responsibility and may change it. The study measured bounded passage selection, not fully autonomous reply quality.[^ref-894d3703de86]
Neighborhood in Abstraction Space¶
Textual Case Based Reasoning sits in a sparse region of the domain-specific corpus (97th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2721 abstractions)
Nearest neighbors
- Example-First Modeling — 0.77
- Extended Boolean model — 0.76
- Uniform Resource Identifier — 0.76
- Text mining — 0.76
- YARA — 0.76
Computed from structural-signature embeddings · 2026-10-17
Not to Be Confused With¶
- Generic document search: topical text matches without a defined case-solving task and prior-case relation.[^ref-4ce3bc7a5903]
- Structured case-based reasoning: prior-case use with no text-bearing knowledge.[^ref-4ce3bc7a5903]
- Legal precedent: prior decisions can carry binding or strongly persuasive authority; that obligation is not required in the email or other textual CBR methods.[ref-4eebad34b0e0][ref-894d3703de86]
- A mandatory four-step CBR cycle: retrieval-focused TCBR can qualify without every query revising and retaining a new case.[^ref-4ce3bc7a5903]
- One system architecture: SMILE+IBP and the email approach are particular implementations, not universal TCBR requirements.[ref-4eebad34b0e0][ref-894d3703de86]
References¶
[^ref-4ce3bc7a5903]: Weber, Ashley, and Brüninghaus, “Textual case-based reasoning,” The Knowledge Engineering Review 20, no. 3 (2005 issue; online 2006), 255–260, DOI 10.1017/S0269888906000713. Publisher record and abstract; full original-author manuscript, §§1 and 3. Commentary/survey used for terminology and boundary, not as an original experiment; manuscript publication metadata is provisional. [^ref-4eebad34b0e0]: Brüninghaus and Ashley (2006), “Progress in Textual Case-Based Reasoning: Predicting the Outcome of Legal Cases from Text,” Proceedings of AAAI 2006, 1577–1580, PDF pp. 1–4, especially SMILE+IBP Figures 1–3 and the evaluation. Original publisher paper. [^ref-894d3703de86]: Lamontagne and Lapalme (2004), “Textual Reuse for Email Response,” Advances in Case-Based Reasoning, LNAI 3155, 242–256, DOI 10.1007/978-3-540-28631-8_19, §§2–6 and Tables 1–2. Complete author-uploaded original paper. The university-hosted PDF was inaccessible on direct open; the author-uploaded original text was inspected.