The Concept
In AI translation, the default objective is semantic integrity — preserving the meaning of the source text. In high-stakes Patent Law, meaning is insufficient. The translation must also adhere to a rigid structural architecture that dictates how that meaning is presented.
Structural compliance refers to the non-semantic, morpho-syntactic frameworks required by statute or local patent practice. These are not stylistic preferences — they are binary constraints.
"A translation that is semantically accurate but structurally non-compliant — e.g., using an active verb where a noun is mandated — can render a patent claim invalid on its face."
The Challenge: When AI Resists the Law
Generic LLMs are trained on the "average internet," where linguistic naturalness and conversational flow are rewarded. Consequently, they aggressively resist the artificial, repetitive, and disjointed syntax of "Patentese."
The Conflict
| What the Model Wants |
What the Law Demands |
| Conjugate verbs to create flow |
Nominalize verbs to define static scope |
| Use natural articles ("a," "the") fluidly |
Enforce rigid definite/indefinite distinction |
| Use de generically or du for flow |
Distinguish Class (de) vs. Instance (du) |
| Translate tokens left-to-right (Head-Final) |
Invert to Head-Initial French noun phrase structure |
Why This Axis Is Auditable
These constraints share one property that makes them the strongest ground for quality measurement: they are binary. A claim either preserves antecedent basis or it does not; a method step is nominalized or it is not. An audit on this axis produces findings, not opinions — every violation can be shown, sourced and counted. It is also where fluency bias does the most silent damage: the model's statistical preference for natural French is precisely what pushes it out of compliance.
The Four Critical Constraints
Constraint 1
Verb Nominalization
Method Claims with "comprising"
The Rule: In French patent practice, when a method claim uses the transition comprising (comprenant), subsequent steps must be nominalized (turned into nouns), not left as infinitive verbs.
Why It Matters: Patent claims define a static scope of protection. Infinitive verbs suggest ongoing actions; nominalized forms define discrete structural elements.
❌ INCORRECT (Fluent but Invalid):
"Procédé comprenant : fixer un élément..."
(Method comprising: to fix an element...)
✓ CORRECT (Structurally Compliant):
"Procédé comprenant : la fixation d'un élément..."
(Method comprising: the fixing of an element...)
Why Models Fail: Infinitive constructions are statistically more common in non-patent French — fluency wins over formalism.
Detection: The trigger is lexical. Every comprenant opens a scope in which verb forms can be checked mechanically, including grammatical parallelism across multi-step lists.
→ View Constraint 1 Case Catalog
Constraint 2
Definite / Indefinite Article Distinction
Antecedent Basis & Referential Integrity
In general linguistics, the choice between "A" (Indefinite) and "The" (Definite) is often a matter of style. In Intellectual Property, it is a matter of strict legal logic known as Antecedent Basis.
The "Closed Loop" Logic
A patent claim functions as a rigorous closed loop. The rules are binary:
- Introduction ("A" / "An"): Must introduce a new element for the first time.
- Reference ("The" / "Said"): Must reference an element already defined.
The AI Blind Spot
Models treat articles as low-value tokens. When faced with complex quantifiers like "of the at least some," they prioritize linguistic fluidity and default to the smoother — but legally fatal — indefinite structure "d'au moins certaines," effectively deleting the definitive pointer and introducing New Matter.
Detection: Fully rule-checkable. Every definite reference in the target must map to an introduction earlier in the claim set — a dependency that can be reconstructed and verified segment by segment.
→ View Constraint 2 Case Catalog
Constraint 3
Genitive Form Differentiation
Class vs. Instance — Prepositional Scope
The Challenge: Unlike the binary rules of verbs, genitive forms exist in a "Gray Zone." The choice of preposition (de, du, or Zero-Marker) defines the engineering relationship between components. AI output typically fails in two opposite directions.
Failure Mode A — Over-Specification (Definiteness Bias)
Source: "Device handle" (Generic type)
❌ AI Output: "Poignée du dispositif"
Verdict: Implies a specific device exists when it doesn't (False Antecedent).
Failure Mode B — Under-Grammaticization (Anglicism Trap)
Source: "Target chamber" (Functional housing)
❌ AI Output: "Chambre cible" (Juxtaposition)
✓ Correct: "Chambre de cible" (Functional marker)
Detection: Because this constraint depends on intent (is it a label or a function?), it cannot be fully rule-checked. Known functional classes are flagged automatically; novel compounds are routed to expert review. The taxonomy records which of the two failure directions each case belongs to.
→ View Constraint 3 Case Catalog
Constraint 4
Syntactic Linearity Bias
Complex Noun Phrases & Modification Scope
The Rule: English technical nomenclature is Head-Final (Modifiers + Noun), whereas French is Head-Initial (Noun + Modifiers). AI models process tokens sequentially without performing the necessary syntactic inversion, disconnecting the noun from its modifiers and creating legally indefinite claims.
Source: "Electrocardiogram, ECG, Lead" (single component)
❌ Raw AI Output (Linear):
"Électrocardiogramme, ECG, Dérivation"
Verdict: Three separate items — a machine, an acronym, a wire.
✓ Expert Correction (Head-Initial Inversion):
"Dérivation d'électrocardiogramme (ECG)"
Verdict: Single legal object correctly defined.
Why Models Fail: "Linearity Bias" pushes the model to output tokens in the exact English order, producing flat lists rather than syntactically valid French compounds.
Detection: Compare the modification structure of source and target noun phrases — a flat, token-order match on a complex compound is the signature of the defect.
→ View Constraint 4 Case Catalog