Norwegian Bokmal stemmer benchmarks ★★★★★

This page reports same-language stemming benchmarks for Norwegian Bokmal. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.

Dictionary size: ★★★★★. The exact count is 73,170 distinct usable word forms after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all benchmarked dictionaries in five nearly equal groups; they do not measure linguistic quality or benchmark accuracy.

All speed values are environment-specific and were measured on the hardware and JVM listed in the benchmark overview. The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java 4.4.0 snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.

How to read this page

Start with the corpus and patch-command distribution, then compare exact-root agreement with runtime. The dictionary-family experiment, edit-cost experiment, and pairwise linguistic evaluation answer separate questions. Their 10–90% curves use independent frozen protocols and must not be substituted for one another.

Runtime and exact-root agreement measure different properties. Light, minimal, possessive, and other rule-based filters intentionally have different transformation scopes, so a lower runtime can coexist with lower dictionary-root agreement. Read the speed and accuracy tables together. The Radixor rows in this refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The EnglishRadixorDictionaryCoverageBenchmark shows the resulting quality/speed envelope explicitly.

Dictionary corpus

Model ID	Model version	Language	Dictionary rows	Distinct usable forms	Complete quality tokens	Already-root tokens	Changed tokens	Timing workload	JMH timing tokens
nb-no-default	1.0.0	NB_NO	17,929	73,170	90,757	33,376	57,381	changed tokens	57,381

Radixor patch-command distribution

Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete default-model dictionary. The total number of preferred patch commands analyzed for this language is 90,757.

Command class	Meaning	Word forms	Share
AppendCharacterCommand	Appends one character to the end of the word form.	2,528	2.785%
BackwardCompoundCommand	Applies a multi-step backward patch made from skip, delete, insert, and replace operations.	4,258	4.692%
DeleteSuffixCommand	Deletes one or more trailing characters from the word form.	48,925	53.908%
PreserveCommand	Returns the word form unchanged because it already matches the preferred root.	32,086	35.354%
ReplaceLastCharacterCommand	Replaces the final character of the word form.	2,960	3.261%

Accuracy

Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.

Stemmer	All exact	Changed exact	Root preserved	Note
Radixor	96.852%	97.637%	95.503%	Radixor dictionary-trained patch-command stemmer.
Lucene NorwegianMinimalStemFilter	57.107%	53.913%	62.599%	Minimal suffix reducer; narrow baseline, not a full stemmer.
Official Snowball direct (Java)	54.824%	51.791%	60.040%	Official Snowball generated Java stemmer; rule-based suffix algorithm.
Lucene SnowballFilter	54.803%	51.780%	60.001%	Lucene TokenFilter integration path around the Snowball algorithm.
Lucene NorwegianLightStemFilter	52.136%	50.616%	54.749%	Light suffix stemmer; intentionally narrower than Radixor's dictionary-trained transformation model.

Speed

Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.

Stemmer	Benchmark method	Score ms/op	Error ms	ns/token	Relative vs Radixor	Note
Radixor	radixor[nb-no-default]	4.085	0.604	71.2	1.000	Radixor dictionary-trained patch-command stemmer.
Lucene NorwegianMinimalStemFilter	norwegianBokmalLuceneNorwegianMinimalStemFilter	2.890	0.168	50.4	0.708	Minimal Norwegian suffix reducer.
Lucene NorwegianLightStemFilter	norwegianBokmalLuceneNorwegianLightStemFilter	3.236	0.061	56.4	0.792	Light Norwegian suffix stemmer.
Official Snowball direct (Java)	snowballDirect[NORWEGIAN_BOKMAL]	5.649	0.720	98.5	1.383	Official Snowball generated Java stemmer; direct API.
Lucene SnowballFilter	luceneSnowballFilter[NORWEGIAN_BOKMAL]	7.283	1.207	126.9	1.783	Lucene TokenFilter path around Snowball; includes TokenStream overhead.

Interpretation notes

• Radixor is a dictionary-trained patch-command stemmer. Its learned transformations can generalize beyond the word forms listed in the training resource.
• Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
• Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
• Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
• Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.

Dictionary-family generalization conclusion

This is the language-specific conclusion from the independent radixor-generalization-v1 baseline experiment. It is intentionally separate from the wider edit-cost protocol below; values from the two frozen snapshots are not substituted for one another.

Evidence

Model nb-no-default version 1.0.0 is evaluated over five predeclared nested splits. Unseen metrics remove withheld occurrences whose normalized surface also appeared in training. Parentheses show the observed split minimum–maximum.

Training rows	Median unseen occurrences	Unseen all exact	Unseen changed exact	Unseen root preserved
10%	81,276	69.855% (69.368–70.692)	58.598% (57.048–59.496)	89.971% (89.224–90.585)
20%	71,829	72.851% (72.564–73.423)	62.099% (61.353–62.757)	91.669% (91.337–91.848)
30%	62,496	74.899% (74.255–75.323)	64.655% (63.730–65.508)	92.387% (92.185–92.553)
40%	53,261	76.064% (75.979–76.622)	66.133% (65.907–66.950)	93.222% (92.859–93.532)
50%	44,080	77.213% (76.853–77.697)	67.372% (67.269–68.071)	94.168% (93.324–94.348)
60%	35,053	77.888% (77.692–78.230)	68.373% (67.934–68.693)	94.544% (94.086–94.654)
70%	26,235	78.616% (78.247–79.117)	69.080% (68.975–69.623)	95.083% (94.075–95.571)
80%	17,357	79.282% (78.391–79.483)	69.935% (68.919–70.124)	95.430% (94.759–95.606)
90%	8,682	79.616% (79.365–80.300)	70.546% (70.229–71.560)	95.510% (95.252–96.117)

Generalization conclusion

• Median exactness on genuinely unseen changed forms moves from 58.598% at 10% training knowledge to 70.546% at 90%, a measured +11.949 percentage-point change.
• Unseen all-form exactness moves from 69.855% at 10% training knowledge to 79.616% at 90%, a measured +9.761 percentage-point change.
• Preservation of unseen already-root forms moves from 89.971% at 10% training knowledge to 95.510% at 90%, a measured +5.538 percentage-point change.
• The evidence establishes within-resource transfer across withheld dictionary families. It does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.

The complete ten-level table and split ranges remain in the complete generalization appendix; raw counters and provenance are in active machine-readable snapshot. The frozen methodology defines family-level splitting, unseen-surface leakage control, aggregation, and the limits of the claim.

Edit costs and dictionary-knowledge generalization

This section interprets the edit-cost and held-out-family experiment for NB_NO separately from the cross-language macro summary. Each knowledge point is the median of five frozen, nested splits. The primary exactness outcome covers changed forms in withheld families after excluding normalized surfaces seen in training. Thus the complete dictionary is the evaluation population, while only genuinely unseen surfaces contribute to this outcome.

Cost labels have the fixed form D<delete>I<insert>R<replace>M<match>. D is the cost of deleting a source character, I of inserting a target character, R of replacing a source character, and M of keeping an equal source/target character unchanged (the match or skip step). For example, D2I5R3M0 means delete cost 2, insert cost 5, replace cost 3, and match cost 0. The numbers are relative dynamic-programming costs, not command counts.

Evidence

Dictionary rows	Evaluated forms	Changed-form share	Baseline commands	Exact cost classes	Grid reduction	Largest exact class
17,929	90,757	63.22%	325	14	16.71×	54

The exact classes are based on command-by-command equality over the complete dictionary, not equality of aggregate trie metrics. A higher class count means that this dictionary exposes more cost-dependent encoder decisions; it does not by itself mean better quality.

Knowledge	Baseline unseen changed exact	Selected-cost exact	Δ	Baseline F0.5	Selected F0.5	Baseline commands	Selected commands
10%	58.507%	58.507%	+0.000 pp	0.8288	0.8288	1.000×	0.974×
20%	61.982%	61.982%	+0.000 pp	0.8489	0.8489	1.000×	0.973×
30%	64.407%	64.407%	+0.000 pp	0.8602	0.8602	1.000×	0.984×
40%	65.976%	65.976%	+0.000 pp	0.8672	0.8672	1.000×	0.962×
50%	67.148%	67.144%	-0.004 pp	0.8726	0.8726	1.000×	0.964×
60%	68.394%	68.394%	+0.000 pp	0.8796	0.8796	1.000×	0.966×
70%	69.404%	69.404%	+0.000 pp	0.8845	0.8845	1.000×	0.966×
80%	70.284%	70.284%	+0.000 pp	0.8883	0.8883	1.000×	0.964×
90%	70.574%	70.574%	+0.000 pp	0.8917	0.8917	1.000×	0.962×

Within-language associations

Spearman coefficients are calculated independently inside each seed × knowledge stratum across the normalized cost grid. The table reports the median and central 95% empirical interval across up to 45 strata. A relationship is called stable only when it is defined in all 45 strata and the interval retains one sign. These intervals are descriptive, not multiplicity-adjusted confidence intervals. Every predictor and outcome label is defined in the methodology glossary.

The strongest structural pairs whose central interval retains one sign are:

Predictor	Structural outcome	Median Spearman ρ	Central 95%	Strata
patch_command_ratio	trie_nodes	+0.992	+0.909…+1.000	45
patch_command_ratio	value_references	+0.992	+0.909…+1.000	45
replace_to_delete_insert	patch_command_ratio	-0.899	-0.905…-0.866	45
replace_to_delete_insert	trie_nodes	-0.902	-0.932…-0.785	45
replace_to_delete_insert	value_references	-0.902	-0.931…-0.785	45
replace_cost	patch_command_ratio	-0.685	-0.741…-0.647	45

For each quality outcome, the largest absolute median association is shown even when its interval crosses zero. This prevents a large median in heterogeneous strata from being misreported as a portable language-level effect.

Predictor	Quality outcome	Median Spearman ρ	Central 95%	Stable	Defined strata
trie_edges	unseen_changed_exact	-1.000	-1.000…-0.840	no	5 / 45
trie_edges	unseen_f05	-1.000	-1.000…+0.165	no	6 / 45
trie_edges	unseen_over_percent	-1.000	-1.000…-1.000	no	3 / 45
trie_edges	unseen_under_percent	+1.000	-0.165…+1.000	no	6 / 45

Edit-cost conclusion

• With baseline costs, median unseen changed-form exactness changes from 58.507% at 10% knowledge to 70.574% at 90%, a +12.068 pp measured knowledge effect.
• The predeclared selection is D1I1R10M0. Its median unseen changed-form exactness differs from baseline by -0.004 pp and it reduces the median retained-command count by 3.54% (0.965× baseline).
• Under the selected costs, the 10%–90% knowledge change is +12.068 pp. This quantifies generalization for this dictionary; it is not a claim about unrelated domains or lexical resources.
• The non-baseline setting is an efficiency candidate, not a production default: it was selected and evaluated on the same matrix and therefore requires external-corpus or external-dictionary validation before adoption.
• No cost or representation predictor is both defined in all 45 strata and retains one association sign over the central 95% interval for an unseen-form quality outcome. Effects with partial coverage are insufficient for a stable language-level claim; the remaining measured effects are heterogeneous across knowledge levels and splits.

The complete evidence is available in the raw logical matrix, the per-language knowledge curves, and the per-language association table. See the cross-language analysis and frozen methodology for scope and limitations.

Stemming quality

Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language NB_NO using the complete validated stemming-quality result matrix and the canonical linguistic-quality methodology, including its overlapping-group relation, dictionary modes, output policies, metrics, and ranking rules.

Radixor's model was trained from the same lexical resource that defines this benchmark's dictionary-reference relation. The result therefore measures same-resource agreement, not independent external linguistic validity. Third-party adapters receive the same evaluated forms but were not trained by this benchmark.

The tables retain both canonical dictionary-processing modes and every applicable output policy without redefining them on each language page. Download the complete machine-readable result snapshot.

Evaluation scope and key findings

The default model is nb-no-default, loaded from classpath resource org/egothor/stemmer/models/nb-no-default/stemmer.gz. The following findings compare only deterministic PRIMARY_OUTPUT rows over identical included groups; candidate policies are reported separately as capability analyses.

• ALL_WORDS: Radixor ranks first by balanced accuracy at 0.976021 among 5 deterministic stemmers. The runner-up is SNOWBALL NORWEGIAN BOKMAL DIRECT at 0.874259, a difference of 0.101762. This rank does not imply leadership in throughput or every secondary metric.
• LOWERCASE_GROUPS_ONLY: Radixor ranks first by balanced accuracy at 0.976240 among 5 deterministic stemmers. The runner-up is SNOWBALL NORWEGIAN BOKMAL DIRECT at 0.874286, a difference of 0.101954. This rank does not imply leadership in throughput or every secondary metric.

ALL_WORDS

This mode contains 7 result rows, 5 evaluated stemmers, and 3 output policies. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Where at least two deterministic rows are available, PRIMARY_OUTPUT and ALL_CANDIDATES rankings are ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. ANY_CANDIDATE has no single rank metric and is listed alphabetically. Balanced accuracy is a navigation metric, not a universally authoritative quality score.

PRIMARY_OUTPUT ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	0.976021	0.000000%	4.795770%
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	0.874259	0.000386%	25.147805%
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	0.874138	0.000389%	25.171937%
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	0.849389	0.000416%	30.121722%
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	0.831282	0.000110%	33.743568%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	1.000000	0.952042	1.000000	0.976021	0.999997	0.000003
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	0.910734	0.748522	0.999996	0.874259	0.999983	0.000017
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	0.910189	0.748281	0.999996	0.874138	0.999983	0.000017
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	0.898501	0.698783	0.999996	0.849389	0.999980	0.000020
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.969387	0.662564	0.999999	0.831282	0.999981	0.000019

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.990026	0.975432	0.961262	0.952042	0.975727	0.975725
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	0.872901	0.821699	0.776171	0.697359	0.825654	0.825646
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	0.872435	0.821332	0.775884	0.696830	0.825274	0.825266
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	0.849918	0.786156	0.731293	0.647658	0.792374	0.792365
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.887216	0.787133	0.707341	0.648985	0.801425	0.801417

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	134138	0	6757	2676746970	0 / 2676746970	6757 / 140895
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	105463	10337	35432	2676736633	10337 / 2676746970	35432 / 140895
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	105429	10403	35466	2676736567	10403 / 2676746970	35466 / 140895
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	98455	11122	42440	2676735848	11122 / 2676746970	42440 / 140895
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	93352	2948	47543	2676744022	2948 / 2676746970	47543 / 140895

ANY_CANDIDATE oracle bounds

These results are measured, not missing. ANY_CANDIDATE answers two separate optimistic questions for each pair: a gold-related pair avoids under-stemming when the candidate sets intersect, while a gold-negative pair avoids over-stemming when some non-colliding candidate selection exists. The oracle may choose a different candidate for the same word in different pairs. Consequently, these decisions do not form one globally realizable predicted relation or one TP/FP/FN/TN confusion matrix. Balanced accuracy, F-scores, Jaccard, Fowlkes–Mallows, and MCC are therefore mathematically not applicable, rather than unknown.

Stemmer	Optimistic over-stemming (OI)	Optimistic under-stemming (UI)
Radixor	0.000000%	0.000000%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
Radixor	0 / 2676746970	0 / 140895

ALL_CANDIDATES result (no same-language comparator)

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
n/a	Radixor	1.000000	0.000000%	0.000000%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
n/a	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	0.000000

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
n/a	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
n/a	Radixor	ALL_CANDIDATES	140895	0	0	2676746970	0 / 2676746970	0 / 140895

Multi-output analysis

Alternative candidates are capability analyses, not replacements for the deterministic comparison.

Stemmer	Under pairs repaired	Best-case over pairs avoided	All-candidate collisions added	Multi-candidate forms	Multi-candidate share	Maximum candidates	Total candidate assignments
Radixor	6757	0	0	2097	2.865929%	9	75343

LOWERCASE_GROUPS_ONLY

This mode contains 7 result rows, 5 evaluated stemmers, and 3 output policies. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Where at least two deterministic rows are available, PRIMARY_OUTPUT and ALL_CANDIDATES rankings are ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. ANY_CANDIDATE has no single rank metric and is listed alphabetically. Balanced accuracy is a navigation metric, not a universally authoritative quality score.

PRIMARY_OUTPUT ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	0.976240	0.000000%	4.751928%
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	0.874286	0.000387%	25.142395%
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	0.874101	0.000387%	25.179325%
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	0.849330	0.000414%	30.133659%
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	0.831210	0.000108%	33.757794%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	1.000000	0.952481	1.000000	0.976240	0.999997	0.000003
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	0.910689	0.748576	0.999996	0.874286	0.999983	0.000017
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	0.910546	0.748207	0.999996	0.874101	0.999983	0.000017
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	0.898862	0.698663	0.999996	0.849330	0.999980	0.000020
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.969896	0.662422	0.999999	0.831210	0.999981	0.000019

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.990121	0.975662	0.961620	0.952481	0.975951	0.975950
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	0.872882	0.821713	0.776211	0.697379	0.825663	0.825655
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	0.872677	0.821432	0.775872	0.696975	0.825395	0.825387
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	0.850142	0.786219	0.731236	0.647743	0.792466	0.792457
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.887506	0.787200	0.707265	0.649077	0.801549	0.801541

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	134115	0	6691	2672431799	0 / 2672431799	6691 / 140806
2	SNOWBALL NORWEGIAN BOKMAL DIRECT	PRIMARY_OUTPUT	105404	10337	35402	2672421462	10337 / 2672431799	35402 / 140806
3	SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER	PRIMARY_OUTPUT	105352	10350	35454	2672421449	10350 / 2672431799	35454 / 140806
4	NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER	PRIMARY_OUTPUT	98376	11069	42430	2672420730	11069 / 2672431799	42430 / 140806
5	NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER	PRIMARY_OUTPUT	93273	2895	47533	2672428904	2895 / 2672431799	47533 / 140806

ANY_CANDIDATE oracle bounds

These results are measured, not missing. ANY_CANDIDATE answers two separate optimistic questions for each pair: a gold-related pair avoids under-stemming when the candidate sets intersect, while a gold-negative pair avoids over-stemming when some non-colliding candidate selection exists. The oracle may choose a different candidate for the same word in different pairs. Consequently, these decisions do not form one globally realizable predicted relation or one TP/FP/FN/TN confusion matrix. Balanced accuracy, F-scores, Jaccard, Fowlkes–Mallows, and MCC are therefore mathematically not applicable, rather than unknown.

Stemmer	Optimistic over-stemming (OI)	Optimistic under-stemming (UI)
Radixor	0.000000%	0.000000%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
Radixor	0 / 2672431799	0 / 140806

ALL_CANDIDATES result (no same-language comparator)

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
n/a	Radixor	1.000000	0.000000%	0.000000%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
n/a	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	0.000000

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
n/a	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
n/a	Radixor	ALL_CANDIDATES	140806	0	0	2672431799	0 / 2672431799	0 / 140806

Multi-output analysis

Alternative candidates are capability analyses, not replacements for the deterministic comparison.

Stemmer	Under pairs repaired	Best-case over pairs avoided	All-candidate collisions added	Multi-candidate forms	Multi-candidate share	Maximum candidates	Total candidate assignments
Radixor	6691	0	0	2064	2.823105%	9	75251

Method and interpretation boundaries

The linguistic-quality methodology defines the overlapping gold relation, both dictionary-processing modes, all three output policies, confusion counts, formulas, undefined denominators, aggregation, ranking, and inapplicable partition metrics. The candidate-policy reference explains why ANY_CANDIDATE is an oracle-assisted capability bound rather than deterministic runtime behavior.

This page preserves the language-specific raw counts, metrics, candidate distributions, comparison availability, and caveats. The machine-readable CSV remains authoritative for every field.

Provenance

• Authoritative source: docs/benchmarks/data/stemming-quality-2026-09-13.csv
• Source SHA-256: d41e00160cda44758e806c37d210f2d9b90b1ebad22eebe5ece89a40f4ed9ab0
• Evaluation command: ./gradlew stemmingQuality --no-daemon
• Dictionary language: NB_NO
• Processing modes: ALL_WORDS, LOWERCASE_GROUPS_ONLY
• Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and gradle.lockfile
• Model ID, version, and SHA-256: recorded in every CSV row
• Run date, core source state, JDK, operating system, and hardware: recorded on the benchmark environment page