Romanet Parsing

The core challenge of romanet parsing is distinguishing between sibling and child elements in deeply nested outlines. A parser must recognize that 'subsection (i)' and 'subsection (ii)' are siblings under a parent clause, while 'sub-subsection (i)' under subsection (ii) begins a new hierarchical level. This requires stateful tracking of indentation, preceding markers, and context.

Romanet parsing converts string representations into their integer equivalents for computational processing. Key rules include:

• Additive principle: 'vi' = 5 + 1 = 6
• Subtractive principle: 'iv' = 5 - 1 = 4
• Valid character set: I, V, X, L, C, D, M only
• Maximum sequential repeats: Three identical symbols (e.g., 'iii' = 3, but 'iv' for 4) This normalization enables sorting, comparison, and cross-referencing with other numbering schemes.

A critical preprocessing step is distinguishing romanet numerals from identical alphabetic text. The string 'i' could be a romanet numeral, the English first-person pronoun, or part of a word like 'in'. Disambiguation strategies include:

• Positional analysis: Is the token at the start of a line or after whitespace?
• Contextual cues: Is it followed by a period, parenthesis, or other delimiter?
• Dictionary exclusion: Is the token a known word in the document's language?
• Pattern matching: Does it conform to a known outline numbering sequence?

Romanet parsing is rarely performed in isolation. It typically operates within a larger outline reconstruction pipeline that handles mixed numbering schemes:

• Level 1: Uppercase Roman (I, II, III)
• Level 2: Uppercase Alpha (A, B, C)
• Level 3: Arabic Numeric (1, 2, 3)
• Level 4: Lowercase Romanet (i, ii, iii) — the romanet layer
• Level 5: Lowercase Alpha (a, b, c) The parser must correctly assign each romanet token to its appropriate depth in the document tree.

Real-world legal documents present numerous parsing challenges:

• Malformed numerals: 'iiii' instead of 'iv' (common in older texts)
• Mixed case: 'Iv' or 'iV' due to OCR errors
• Overlapping ranges: Romanet sequences that restart within sibling sections
• Non-standard delimiters: Periods, double parentheses, or em-dashes
• Inline references: 'as described in subsection (iv) above' Robust parsers must handle these variations gracefully without breaking the structural parse.

Romanet parsing serves as a foundational preprocessing step for downstream legal AI tasks:

• Citation resolution: Linking 'see § 5(a)(iv)' to the correct target
• Clause extraction: Isolating operative provisions by their romanet identifiers
• Document comparison: Aligning corresponding sub-clauses across versions
• Obligation extraction: Identifying duties tied to specific sub-provisions
• Summarization: Preserving hierarchical context in condensed outputs Accurate romanet parsing directly impacts the citation integrity of legal reasoning systems.

The process of identifying section titles and subtitles and reconstructing the nested parent-child relationships that form the document's outline. Romanet parsing is a specialized subset of this broader task.

• Detects font size, weight, and indentation as structural signals
• Reconstructs the logical tree of sections, subsections, and clauses
• Handles mixed numbering schemes: Arabic, Roman, and alphabetical
• Critical for generating accurate table of contents and navigation panes

Example: A contract may use 'Article I' (uppercase Roman), 'Section 1.1' (Arabic decimal), and '(a)' (lowercase alpha) simultaneously. Header hierarchy extraction must normalize all three into a coherent outline.

The specialized task of decomposing a citation to a statute into its constituent parts, including title, chapter, section, and subsection numbers. Romanet values frequently appear as subsection identifiers within these references.

• Parses strings like '42 U.S.C. § 1983(b)(iii)' into structured fields
• Must distinguish romanet 'iii' from the word 'iii' in party names
• Enables cross-reference resolution and hyperlink generation
• Validates citations against authority databases for correctness

Statutory parsers often use the same romanet normalization logic to convert '§ 5(a)(iv)' into a machine-readable path for retrieval systems.

A token-level annotation standard used in sequence labeling where tokens are tagged as the Beginning, Inside, or Outside of a named entity or structural chunk. This is the dominant approach for training romanet detection models.

• B-Romanet: Marks the first token of a romanet sequence
• I-Romanet: Marks continuation tokens within the same romanet
• O: Marks tokens outside any romanet span
• Enables token classification models to learn structural boundaries

Training data example: 'subsection ( B-Romanet iv I-Romanet ) O shall O govern O' teaches the model to recognize parenthetical romanet spans as distinct from surrounding legal text.

The process of computationally linking a textual reference pointer within a legal document to the specific target provision, section, or external authority it cites. Romanet parsing is a prerequisite for resolving internal cross-references.

• Resolves 'as set forth in § 2(a)(iv) above' to the exact clause
• Requires normalized identifiers — romanets must be converted to a canonical form
• Handles Id. references that point to the immediately preceding authority
• Builds a citation graph for network analysis of legal reasoning

Without accurate romanet parsing, a reference to 'subsection (v)' cannot be reliably matched to its target, breaking the citation network that powers legal research tools.

The task of assigning a functional label — such as 'title', 'recital', 'operative provision', or 'signature block' — to a segmented block of text within a legal document. Romanet-labeled blocks typically receive the 'operative provision' classification.

• Distinguishes binding clauses from prefatory recitals
• Uses spatial features (indentation, position) and textual features (numbering patterns)
• Romanet presence is a strong signal that text is a subordinate operative clause
• Enables downstream tasks like obligation extraction and clause comparison

A block beginning with '(iii)' is almost certainly a binding sub-provision, not a recital or definition. This classification signal helps legal AI systems prioritize which text to analyze for deontic logic.

A rule-based method for inferring document structure by analyzing changes in font size, weight, and style to detect headings and hierarchy. Romanet paragraphs often share identical font characteristics with their parent clause, making them invisible to font-based heuristics alone.

• Detects heading levels by font size thresholds
• Identifies bold, italic, and small caps as emphasis signals
• Limitation: Cannot distinguish romanet '(i)' from body text if font is unchanged
• Must be combined with regex-based numbering detection for complete coverage

This limitation is precisely why dedicated romanet parsing is necessary — font heuristics fail when the only structural signal is the numbering scheme itself.