Spanish stemmer benchmarks ★★★★★

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

Dictionary size: ★★★★★. The exact count is 849,661 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
es-es-default	1.0.0	ES_ES	65,059	849,661	926,393	120,121	806,272	changed tokens	806,272

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 926,393.

Command class	Meaning	Word forms	Share
AppendCharacterCommand	Appends one character to the end of the word form.	8,534	0.921%
BackwardCompoundCommand	Applies a multi-step backward patch made from skip, delete, insert, and replace operations.	522,685	56.422%
DeleteSuffixCommand	Deletes one or more trailing characters from the word form.	243,410	26.275%
PreserveCommand	Returns the word form unchanged because it already matches the preferred root.	124,386	13.427%
ReplaceLastCharacterCommand	Replaces the final character of the word form.	27,378	2.955%

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	97.459%	97.544%	96.891%	Radixor dictionary-trained patch-command stemmer.
Lucene HunspellStemFilter	49.074%	42.656%	92.154%	Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter.
Lucene SpanishMinimalStemFilter	17.284%	5.347%	97.403%	Minimal suffix reducer; narrow baseline, not a full stemmer.
Lucene SpanishPluralStemFilter	15.140%	5.802%	77.820%	Plural-focused suffix reducer; narrow baseline.
Lucene SpanishLightStemFilter	9.577%	7.088%	26.279%	Light suffix stemmer; intentionally narrower than Radixor's dictionary-trained transformation model.
Lucene SnowballFilter	4.889%	4.287%	8.932%	Lucene TokenFilter integration path around the Snowball algorithm.
Official Snowball direct (Java)	4.889%	4.287%	8.930%	Official Snowball generated Java stemmer; rule-based suffix algorithm.

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[es-es-default]	85.685	15.183	106.3	1.000	Radixor dictionary-trained patch-command stemmer.
Lucene HunspellStemFilter	luceneHunspellStemFilter	2033.994	124.268	2522.7	23.738	Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter.
Lucene SpanishMinimalStemFilter	spanishLuceneSpanishMinimalStemFilter	40.636	2.188	50.4	0.474	Minimal Spanish suffix reducer; narrow baseline.
Lucene SpanishLightStemFilter	spanishLuceneSpanishLightStemFilter	43.418	2.735	53.9	0.507	Light Spanish suffix stemmer.
Lucene SpanishPluralStemFilter	spanishLuceneSpanishPluralStemFilter	93.087	6.142	115.5	1.086	Plural-oriented Spanish suffix reducer.
Official Snowball direct (Java)	snowballDirect[SPANISH]	185.383	13.427	229.9	2.164	Official Snowball generated Java stemmer; direct API.
Lucene SnowballFilter	luceneSnowballFilter[SPANISH]	217.481	19.863	269.7	2.538	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 es-es-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%	827,660	82.066% (81.789–82.910)	80.995% (80.671–81.943)	89.364% (88.856–89.475)
20%	732,142	84.175% (83.996–84.838)	83.310% (83.160–84.044)	90.087% (89.771–90.303)
30%	638,770	85.558% (85.336–86.089)	84.789% (84.495–85.404)	90.925% (90.760–91.204)
40%	544,394	86.496% (86.375–86.920)	85.764% (85.618–86.311)	91.721% (91.215–91.822)
50%	450,843	87.642% (87.181–87.836)	86.991% (86.460–87.243)	92.320% (92.067–92.411)
60%	357,877	88.208% (87.787–88.272)	87.558% (87.076–87.614)	92.915% (92.860–93.031)
70%	265,246	88.730% (88.448–88.971)	88.095% (87.737–88.338)	93.593% (93.457–93.713)
80%	177,896	89.154% (88.946–89.572)	88.448% (88.281–88.984)	93.934% (93.855–94.442)
90%	89,614	89.801% (89.306–90.166)	89.176% (88.546–89.533)	94.711% (94.645–94.920)

Generalization conclusion

• Median exactness on genuinely unseen changed forms moves from 80.995% at 10% training knowledge to 89.176% at 90%, a measured +8.181 percentage-point change.
• Unseen all-form exactness moves from 82.066% at 10% training knowledge to 89.801% at 90%, a measured +7.735 percentage-point change.
• Preservation of unseen already-root forms moves from 89.364% at 10% training knowledge to 94.711% at 90%, a measured +5.348 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 ES_ES 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
65,059	926,393	87.03%	1,496	18	13.00×	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%	80.556%	80.556%	+0.000 pp	0.9362	0.9362	1.000×	0.749×
20%	83.316%	83.316%	+0.000 pp	0.9491	0.9491	1.000×	0.776×
30%	85.068%	85.068%	+0.000 pp	0.9559	0.9559	1.000×	0.795×
40%	86.248%	86.248%	+0.000 pp	0.9599	0.9599	1.000×	0.806×
50%	87.056%	87.056%	+0.000 pp	0.9628	0.9628	1.000×	0.819×
60%	87.683%	87.683%	+0.000 pp	0.9653	0.9653	1.000×	0.827×
70%	88.195%	88.195%	+0.000 pp	0.9665	0.9665	1.000×	0.839×
80%	88.552%	88.552%	+0.000 pp	0.9677	0.9677	1.000×	0.842×
90%	89.280%	89.280%	+0.000 pp	0.9688	0.9688	1.000×	0.851×

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	value_references	+0.993	+0.966…+1.000	45
patch_command_ratio	trie_nodes	+0.953	+0.946…+1.000	45
replace_to_delete_insert	dense_table_slots	+0.947	+0.920…+0.952	45
replace_to_delete_insert	trie_edges	+0.946	+0.867…+0.952	45
replace_to_delete_insert	trie_nodes	+0.867	+0.867…+0.937	45
patch_command_ratio	trie_edges	+0.875	+0.830…+0.978	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
average_path_length	unseen_changed_exact	-1.000	-1.000…-1.000	yes	45 / 45
average_path_length	unseen_f05	-1.000	-1.000…-0.864	yes	45 / 45
average_path_length	unseen_over_percent	-0.821	-1.000…+0.969	no	10 / 45
average_path_length	unseen_under_percent	+1.000	+0.944…+1.000	yes	45 / 45

Edit-cost conclusion

• With baseline costs, median unseen changed-form exactness changes from 80.556% at 10% knowledge to 89.280% at 90%, a +8.725 pp measured knowledge effect.
• The predeclared selection is D10I10R1M1. Its median unseen changed-form exactness differs from baseline by +0.000 pp and it reduces the median retained-command count by 18.02% (0.820× baseline).
• Under the selected costs, the 10%–90% knowledge change is +8.725 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.
• 38 cost/representation-to-quality association(s) are defined in all 45 strata and retain one sign over their central 95% interval. Their direction is evidence for this resource only; inspect the table and machine-readable coefficients before extrapolating.

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 ES_ES 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 es-es-default, loaded from classpath resource org/egothor/stemmer/models/es-es-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 0.989448 among 7 deterministic stemmers. The runner-up is SNOWBALL SPANISH LUCENE FILTER at 0.652438, a difference of 0.337010. This rank does not imply leadership in throughput or every secondary metric.
• LOWERCASE_GROUPS_ONLY: Radixor ranks first by balanced accuracy at 0.989580 among 7 deterministic stemmers. The runner-up is SNOWBALL SPANISH DIRECT at 0.652542, a difference of 0.337038. This rank does not imply leadership in throughput or every secondary metric.

ALL_WORDS

This mode contains 11 result rows, 7 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	0.989448	0.000000%	2.110334%
2	SNOWBALL SPANISH LUCENE FILTER	0.652438	0.000414%	69.511918%
3	SNOWBALL SPANISH DIRECT	0.652438	0.000413%	69.511932%
4	HUNSPELL SPANISH LUCENE FILTER	0.615028	0.000068%	76.994273%
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	0.514565	0.000009%	97.087060%
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	0.503764	0.000002%	99.247265%
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	0.501678	0.000001%	99.664470%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	1.000000	0.978897	1.000000	0.989448	0.999998	0.000002
2	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.895438	0.304881	0.999996	0.652438	0.999915	0.000085
3	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	0.895510	0.304881	0.999996	0.652438	0.999915	0.000085
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.975281	0.230057	0.999999	0.615028	0.999910	0.000090
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	0.974423	0.029129	1.000000	0.514565	0.999887	0.000113
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	0.979154	0.007527	1.000000	0.503764	0.999885	0.000115
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.970596	0.003355	1.000000	0.501678	0.999884	0.000116

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.995707	0.989336	0.983046	0.978897	0.989392	0.989391
2	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.645406	0.454882	0.351206	0.294400	0.522496	0.522469
3	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	0.645436	0.454891	0.351208	0.294407	0.522517	0.522490
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.591847	0.372295	0.271557	0.228724	0.473678	0.473655
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	0.130091	0.056568	0.036142	0.029107	0.168477	0.168467
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	0.036514	0.014940	0.009391	0.007526	0.085851	0.085846
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.016548	0.006687	0.004191	0.003355	0.057067	0.057063

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	41053986	0	885054	360919543590	0 / 360919543590	885054 / 41939040
2	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	12786409	1493087	29152631	360918050503	1493087 / 360919543590	29152631 / 41939040
3	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	12786403	1491944	29152637	360918051646	1491944 / 360919543590	29152637 / 41939040
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	9648381	244539	32290659	360919299051	244539 / 360919543590	32290659 / 41939040
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	1221659	32066	40717381	360919511524	32066 / 360919543590	40717381 / 41939040
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	315690	6721	41623350	360919536869	6721 / 360919543590	41623350 / 41939040
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	140718	4263	41798322	360919539327	4263 / 360919543590	41798322 / 41939040

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)
HUNSPELL SPANISH LUCENE FILTER	0.000062%	76.009935%
Radixor	0.000000%	0.001493%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
HUNSPELL SPANISH LUCENE FILTER	223500 / 360919543590	31877837 / 41939040
Radixor	0 / 360919543590	626 / 41939040

ALL_CANDIDATES ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	0.999993	<0.000001%	0.001493%
2	HUNSPELL SPANISH LUCENE FILTER	0.619950	0.000073%	76.009935%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	ALL_CANDIDATES	0.999959	0.999985	1.000000	0.999993	1.000000	0.000000
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	0.974467	0.239901	0.999999	0.619950	0.999911	0.000089

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	ALL_CANDIDATES	0.999964	0.999972	0.999980	0.999944	0.999972	0.999972
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	0.604361	0.385016	0.282490	0.238402	0.483503	0.483480

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	ALL_CANDIDATES	41938414	1737	626	360919541853	1737 / 360919543590	626 / 41939040
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	10061203	263629	31877837	360919279961	263629 / 360919543590	31877837 / 41939040

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
HUNSPELL SPANISH LUCENE FILTER	412822	21039	19090	11309	1.331001%	5	861853
Radixor	884428	0	1737	20967	2.467690%	21	871404

LOWERCASE_GROUPS_ONLY

This mode contains 11 result rows, 7 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	0.989580	<0.000001%	2.084022%
2	SNOWBALL SPANISH DIRECT	0.652542	0.000410%	69.491126%
3	SNOWBALL SPANISH LUCENE FILTER	0.652542	0.000410%	69.491126%
4	HUNSPELL SPANISH LUCENE FILTER	0.614924	0.000068%	77.015229%
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	0.514575	0.000009%	97.085003%
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	0.503767	0.000002%	99.246590%
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	0.501679	0.000001%	99.664108%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	0.999999	0.979160	1.000000	0.989580	0.999998	0.000002
2	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	0.896551	0.305089	0.999996	0.652542	0.999915	0.000085
3	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.896551	0.305089	0.999996	0.652542	0.999915	0.000085
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.975224	0.229848	0.999999	0.614924	0.999910	0.000090
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	0.974539	0.029150	1.000000	0.514575	0.999887	0.000113
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	0.979521	0.007534	1.000000	0.503767	0.999884	0.000116
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.971230	0.003359	1.000000	0.501679	0.999884	0.000116

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.995761	0.989470	0.983258	0.979159	0.989525	0.989523
2	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	0.646055	0.455257	0.351461	0.294714	0.522999	0.522972
3	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.646055	0.455257	0.351461	0.294714	0.522999	0.522972
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	0.591553	0.372016	0.271323	0.228513	0.473448	0.473426
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	0.130175	0.056607	0.036167	0.029128	0.168546	0.168536
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	0.036546	0.014953	0.009400	0.007533	0.085906	0.085901
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	0.016565	0.006695	0.004195	0.003359	0.057116	0.057113

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	40958710	34	871756	359407144881	34 / 359407144915	871756 / 41830466
2	SNOWBALL SPANISH DIRECT	PRIMARY_OUTPUT	12762004	1472547	29068462	359405672368	1472547 / 359407144915	29068462 / 41830466
3	SNOWBALL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	12762004	1472547	29068462	359405672368	1472547 / 359407144915	29068462 / 41830466
4	HUNSPELL SPANISH LUCENE FILTER	PRIMARY_OUTPUT	9614637	244260	32215829	359406900655	244260 / 359407144915	32215829 / 41830466
5	SPANISH LUCENE SPANISH LIGHT STEM FILTER	PRIMARY_OUTPUT	1219357	31857	40611109	359407113058	31857 / 359407144915	40611109 / 41830466
6	SPANISH LUCENE SPANISH PLURAL STEM FILTER	PRIMARY_OUTPUT	315155	6589	41515311	359407138326	6589 / 359407144915	41515311 / 41830466
7	SPANISH LUCENE SPANISH MINIMAL STEM FILTER	PRIMARY_OUTPUT	140505	4162	41689961	359407140753	4162 / 359407144915	41689961 / 41830466

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)
HUNSPELL SPANISH LUCENE FILTER	0.000062%	76.037484%
Radixor	0.000000%	0.000000%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
HUNSPELL SPANISH LUCENE FILTER	223274 / 359407144915	31806834 / 41830466
Radixor	0 / 359407144915	0 / 41830466

ALL_CANDIDATES ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	1.000000	<0.000001%	0.000000%
2	HUNSPELL SPANISH LUCENE FILTER	0.619812	0.000073%	76.037484%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	ALL_CANDIDATES	0.999987	1.000000	1.000000	1.000000	1.000000	0.000000
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	0.974405	0.239625	0.999999	0.619812	0.999911	0.000089

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	ALL_CANDIDATES	0.999989	0.999993	0.999997	0.999987	0.999993	0.999993
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	0.603992	0.384656	0.282183	0.238126	0.483210	0.483187

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	ALL_CANDIDATES	41830466	560	0	359407144355	560 / 359407144915	0 / 41830466
2	HUNSPELL SPANISH LUCENE FILTER	ALL_CANDIDATES	10023632	263289	31806834	359406881626	263289 / 359407144915	31806834 / 41830466

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
HUNSPELL SPANISH LUCENE FILTER	408995	20986	19029	11287	1.331204%	5	860048
Radixor	871756	34	526	20911	2.466272%	21	869542

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: ES_ES
• 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