CSL Observatory 13 years of Cologne Digital Sanskrit Lexicon

Paradigm-cell coverage

How many finite-verb paradigm cells each DCS root lemma attests — the org-wide version of what the VisualDCS paradigm browser shows on a handful of roots. A cell is a DCS UD 5-tuple (tense · mood · voice · person · number) on finite tokens only.

Roots (≥1 finite)

Finite tokens

≥5 cells

Max cells / root

:::note Trust block. Source: data/paradigm_cell_coverage_per_root.tsv (loader: observatory/site/src/data/paradigm_cell_coverage_per_root.csv.py, read-only). Generator: scripts/paradigm_cell_coverage.py from VisualDCS dcs_full.sqlite. n = roots · finite tokens. Data date: 13-07-2026 (H817 WS1.2). Caveat (B11 / WhitneyRoots): unaccented DCS cannot split class I/VI or IV/passive; Tense=Past conflates aorist/perfect. Cells never overcount traditional lakāra; they can undercount. Non-finite forms (participles, converbs, gerundives, infinitives) are out of scope. :::

Coverage summary

Share of roots clearing cell-count floors — the long tail of hapax finite forms vs paradigm-rich lemmas.

Histogram of distinct cells

How to read: x = number of distinct finite cells per root; y = root count. Example 1: A tall bar at 1 is hapax-cell lemmas (one form only). Example 2: The right tail is high-coverage roots used across many lakāra cells in running text.

Tokens vs cells (log × linear)

How to read: Each dot is a root; x = finite tokens (log), y = distinct cells. Example 1: Points high and left are morphologically diverse for their frequency. Example 2: Points low and right are frequent but paradigm-narrow (few cells).

Top roots by cell coverage

Most frequent cell labels (corpus-wide)

Parse the per-root cells field (Label:count|…) and sum token counts across roots for the top 30 labels.

How to read: Bar length = total finite tokens carrying that cell label. Example 1: Pres.Ind.Act.3.Sing dominates running Sanskrit finite usage. Example 2: Rare duals and passives sit at the long tail.

Conclusion: Finite coverage is heavily skewed: most roots show one or a few cells, while a small head of high-frequency roots open dozens of cells. The cell inventory itself is also skewed toward present indicative active 3sg and a handful of narrative past/optative cells — not a flat traditional paradigm chart.

Data table

Coverage floors (machine-readable summary of the first chart):

Floor Roots Share
≥1 cell %
≥5 cells %
≥20 cells %

Download source TSV: paradigm_cell_coverage_per_root.tsv · report: reports/paradigm_cell_coverage.md · sibling census: L2 Morphology · L5 Roots.