Polish stemmer benchmarks ★★★★★

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

Dictionary size: ★★★★★. The exact count is 120,867 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
pl-pl-unimorph	1.0.0	PL_PL	9,990	120,867	132,308	19,957	112,351	changed tokens	112,351

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 132,308.

Command class	Meaning	Word forms	Share
AppendCharacterCommand	Appends one character to the end of the word form.	1,836	1.388%
BackwardCompoundCommand	Applies a multi-step backward patch made from skip, delete, insert, and replace operations.	52,996	40.055%
DeleteSuffixCommand	Deletes one or more trailing characters from the word form.	37,137	28.069%
PreserveCommand	Returns the word form unchanged because it already matches the preferred root.	20,219	15.282%
ReplaceLastCharacterCommand	Replaces the final character of the word form.	20,120	15.207%

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	98.837%	98.744%	99.359%	Radixor dictionary-trained patch-command stemmer.
Lucene HunspellStemFilter	89.545%	88.272%	96.713%	Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter.
Lucene MorfologikFilter	87.729%	86.606%	94.047%	Dictionary-based path; Morfologik can emit multiple terms.
Lucene StempelFilter	70.009%	69.262%	74.220%	Lucene TokenFilter integration path for table-driven Polish Stempel.
Lucene StempelStemmer direct	70.009%	69.262%	74.220%	Direct table-driven Polish Stempel stemmer API.
Official Snowball direct (Java)	22.315%	20.225%	34.078%	Official Snowball 3.1.0 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[pl-pl-unimorph]	12.165	2.202	108.3	1.000	Radixor dictionary-trained patch-command stemmer.
Lucene HunspellStemFilter	luceneHunspellStemFilter	540.950	39.356	4814.8	44.466	Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter.
Lucene StempelStemmer direct	polishLuceneStempelStemmerDirect	34.399	1.421	306.2	2.828	Direct table-driven Polish Stempel stemmer API.
Lucene StempelFilter	polishLuceneStempelFilter	41.206	1.179	366.8	3.387	Lucene TokenFilter integration path for table-driven Polish Stempel.
Lucene MorfologikFilter	polishLuceneMorfologikFilter	144.144	2.779	1283.0	11.849	Dictionary-based Morfologik TokenFilter; may emit multiple terms.
Official Snowball direct (Java)	snowballDirect[POLISH]	11.946	1.916	106.3	0.982	Official Snowball 3.1.0 generated Java stemmer; direct API. The reported Radixor and Snowball 99.9% JMH intervals overlap; this summary does not support a directional conclusion, and the runtime ratio remains a point estimate.

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.

Optional PoliMorf model

PoliMorf is an independently available optional Polish model and is not the PL_PL default or a member of either standard aggregate. Its evidence is kept separate from the UniMorph default above.

Optional dictionary size: ★★★★★. The exact count is 4,668,685 distinct usable word forms; the rating is relative size, not linguistic quality.

PoliMorf corpus

Model ID	Model version	Language	Dictionary rows	Distinct usable forms	Complete quality tokens	Already-root tokens	Changed tokens	Timing workload	JMH timing tokens
pl-pl-polimorf	1.0.0	PL_PL	315,639	4,668,685	4,812,023	315,639	4,496,384	changed tokens	4,496,384

PoliMorf exact-root accuracy

Stemmer	All exact	Changed exact	Root preserved	Note
Radixor PoliMorf	98.837%	98.744%	99.359%	Exact optional model.
PoliMorf Lucene MorfologikFilter	87.729%	86.606%	94.047%	Matching Polish dictionary adapter.
PoliMorf Official Snowball direct	22.315%	20.225%	34.078%	Matching Polish Snowball algorithm.

PoliMorf runtime timing

Stemmer	Benchmark method	Score ms/op	Error ms	ns/token	Relative vs Radixor PoliMorf	Note
Radixor PoliMorf	radixor[pl-pl-polimorf]	930.102	17.399	206.9	1.000	Exact optional-model baseline.
Lucene MorfologikFilter	polishPolimorfLuceneMorfologikFilter	5912.219	246.399	1314.9	6.357	PoliMorf-derived corpus.

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 pl-pl-unimorph 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%	118,711	65.763% (65.395–67.001)	61.996% (61.861–63.283)	86.633% (85.310–87.965)
20%	105,119	68.340% (67.352–68.619)	64.876% (63.830–65.097)	87.827% (87.200–88.611)
30%	91,501	69.088% (68.299–70.063)	65.677% (64.723–66.909)	88.256% (87.960–88.753)
40%	78,368	70.344% (69.790–71.025)	67.165% (66.449–67.971)	88.216% (88.039–88.588)
50%	65,434	71.160% (70.975–71.928)	68.066% (67.805–69.051)	88.517% (88.330–89.078)
60%	52,228	71.858% (71.219–72.488)	68.868% (68.033–69.513)	89.261% (88.574–89.356)
70%	39,130	72.217% (71.567–72.865)	69.231% (68.375–70.016)	89.428% (88.990–89.729)
80%	25,947	73.298% (71.542–73.560)	70.344% (68.301–70.589)	90.225% (89.430–90.594)
90%	12,963	73.070% (71.833–74.408)	69.972% (68.686–71.530)	90.985% (89.948–91.264)

Generalization conclusion

• Median exactness on genuinely unseen changed forms moves from 61.996% at 10% training knowledge to 69.972% at 90%, a measured +7.976 percentage-point change.
• Unseen all-form exactness moves from 65.763% at 10% training knowledge to 73.070% at 90%, a measured +7.307 percentage-point change.
• Preservation of unseen already-root forms moves from 86.633% at 10% training knowledge to 90.985% at 90%, a measured +4.352 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 PL_PL 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
9,990	132,308	84.92%	846	19	12.32×	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%	61.850%	61.850%	+0.000 pp	0.8594	0.8593	1.000×	0.980×
20%	65.469%	65.469%	+0.000 pp	0.8818	0.8818	1.000×	0.967×
30%	66.483%	66.483%	+0.000 pp	0.8867	0.8867	1.000×	0.951×
40%	67.190%	67.190%	+0.000 pp	0.8926	0.8926	1.000×	0.947×
50%	67.725%	67.725%	+0.000 pp	0.8952	0.8952	1.000×	0.943×
60%	68.669%	68.669%	+0.000 pp	0.8963	0.8963	1.000×	0.939×
70%	69.319%	69.319%	+0.000 pp	0.9031	0.9031	1.000×	0.937×
80%	69.764%	69.764%	+0.000 pp	0.9062	0.9062	1.000×	0.938×
90%	70.138%	70.138%	+0.000 pp	0.9064	0.9064	1.000×	0.935×

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
replace_cost	trie_nodes	+0.777	+0.694…+0.821	45
patch_command_ratio	value_references	+0.954	+0.687…+1.000	45
replace_to_delete_insert	trie_nodes	+0.779	+0.680…+0.858	45
replace_to_delete_insert	trie_edges	+0.732	+0.654…+0.889	45
replace_to_delete_insert	dense_table_slots	+0.733	+0.654…+0.874	45
replace_cost	trie_edges	+0.687	+0.632…+0.796	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…+0.950	no	42 / 45
average_path_length	unseen_f05	-1.000	-1.000…+0.745	no	42 / 45
average_path_length	unseen_over_percent	-0.845	-1.000…+0.397	no	10 / 45
average_path_length	unseen_under_percent	+1.000	-0.955…+1.000	no	42 / 45

Edit-cost conclusion

• With baseline costs, median unseen changed-form exactness changes from 61.850% at 10% knowledge to 70.138% at 90%, a +8.288 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 5.88% (0.941× baseline).
• Under the selected costs, the 10%–90% knowledge change is +8.288 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 PL_PL 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 pl-pl-unimorph, loaded from classpath resource org/egothor/stemmer/models/pl-pl-unimorph/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 0.991105 among 6 deterministic stemmers. The runner-up is POLISH LUCENE MORFOLOGIK FILTER at 0.948392, a difference of 0.042713. This rank does not imply leadership in throughput or every secondary metric.
• LOWERCASE_GROUPS_ONLY: Radixor ranks first by balanced accuracy at 0.991301 among 6 deterministic stemmers. The runner-up is POLISH LUCENE MORFOLOGIK FILTER at 0.948417, a difference of 0.042884. This rank does not imply leadership in throughput or every secondary metric.

ALL_WORDS

This mode contains 12 result rows, 6 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	0.991105	0.000000%	1.779024%
2	POLISH LUCENE MORFOLOGIK FILTER	0.948392	0.001042%	10.320543%
3	HUNSPELL POLISH LUCENE FILTER	0.933457	0.000383%	13.308172%
4	POLISH LUCENE STEMPEL DIRECT	0.855699	0.000602%	28.859618%
5	POLISH LUCENE STEMPEL FILTER	0.855699	0.000602%	28.859618%
6	SNOWBALL POLISH DIRECT	0.823625	0.000967%	35.273970%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	1.000000	0.982210	1.000000	0.991105	0.999997	0.000003
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	0.929398	0.896795	0.999990	0.948392	0.999974	0.000026
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	0.971931	0.866918	0.999996	0.933457	0.999976	0.000024
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	0.947549	0.711404	0.999994	0.855699	0.999950	0.000050
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	0.947549	0.711404	0.999994	0.855699	0.999950	0.000050
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	0.910978	0.647260	0.999990	0.823625	0.999936	0.000064

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.996391	0.991025	0.985717	0.982210	0.991065	0.991064
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	0.922689	0.912805	0.903131	0.839597	0.912951	0.912938
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	0.948942	0.916426	0.886065	0.845744	0.917924	0.917913
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	0.888559	0.812669	0.748723	0.684450	0.821030	0.821007
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	0.888559	0.812669	0.748723	0.684450	0.821030	0.821007
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	0.842338	0.756803	0.687038	0.608756	0.767880	0.767852

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	1097200	0	19873	7303238338	0 / 7303238338	19873 / 1117073
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	1001785	76101	115288	7303162237	76101 / 7303238338	115288 / 1117073
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	968411	27967	148662	7303210371	27967 / 7303238338	148662 / 1117073
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	794690	43990	322383	7303194348	43990 / 7303238338	322383 / 1117073
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	794690	43990	322383	7303194348	43990 / 7303238338	322383 / 1117073
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	723037	70656	394036	7303167682	70656 / 7303238338	394036 / 1117073

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 POLISH LUCENE FILTER	0.000356%	7.227639%
POLISH LUCENE MORFOLOGIK FILTER	0.001000%	2.493123%
Radixor	0.000000%	0.000000%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
HUNSPELL POLISH LUCENE FILTER	25967 / 7303238338	80738 / 1117073
POLISH LUCENE MORFOLOGIK FILTER	73019 / 7303238338	27850 / 1117073
Radixor	0 / 7303238338	0 / 1117073

ALL_CANDIDATES ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	1.000000	0.000000%	0.000000%
2	POLISH LUCENE MORFOLOGIK FILTER	0.987528	0.001376%	2.493123%
3	HUNSPELL POLISH LUCENE FILTER	0.963859	0.000609%	7.227639%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	0.000000
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	0.915516	0.975069	0.999986	0.987528	0.999982	0.000018
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	0.958830	0.927724	0.999994	0.963859	0.999983	0.000017

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	0.926837	0.944354	0.962546	0.894575	0.944823	0.944815
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	0.952443	0.943020	0.933782	0.892184	0.943149	0.943140

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	ALL_CANDIDATES	1117073	0	0	7303238338	0 / 7303238338	0 / 1117073
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	1089223	100514	27850	7303137824	100514 / 7303238338	27850 / 1117073
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	1036335	44498	80738	7303193840	44498 / 7303238338	80738 / 1117073

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 POLISH LUCENE FILTER	67924	2000	16531	10485	8.674824%	6	132492
POLISH LUCENE MORFOLOGIK FILTER	87438	3082	24413	11776	9.742941%	5	133810
Radixor	19873	0	0	1392	1.151679%	4	122430

LOWERCASE_GROUPS_ONLY

This mode contains 12 result rows, 6 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	0.991301	0.000000%	1.739895%
2	POLISH LUCENE MORFOLOGIK FILTER	0.948417	0.001067%	10.315578%
3	HUNSPELL POLISH LUCENE FILTER	0.933546	0.000382%	13.290396%
4	POLISH LUCENE STEMPEL DIRECT	0.856335	0.000611%	28.732387%
5	POLISH LUCENE STEMPEL FILTER	0.856335	0.000611%	28.732387%
6	SNOWBALL POLISH DIRECT	0.823465	0.000990%	35.306102%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	PRIMARY_OUTPUT	1.000000	0.982601	1.000000	0.991301	0.999997	0.000003
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	0.929032	0.896844	0.999989	0.948417	0.999973	0.000027
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	0.972469	0.867096	0.999996	0.933546	0.999975	0.000025
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	0.947796	0.712676	0.999994	0.856335	0.999949	0.000051
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	0.947796	0.712676	0.999994	0.856335	0.999949	0.000051
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	0.910487	0.646939	0.999990	0.823465	0.999935	0.000065

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	PRIMARY_OUTPUT	0.996471	0.991224	0.986032	0.982601	0.991262	0.991261
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	0.922411	0.912654	0.903102	0.839342	0.912796	0.912783
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	0.949394	0.916764	0.886303	0.846320	0.918272	0.918260
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	0.889130	0.813590	0.749881	0.685758	0.821871	0.821848
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	0.889130	0.813590	0.749881	0.685758	0.821871	0.821848
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	0.841894	0.756414	0.686693	0.608253	0.767483	0.767454

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	PRIMARY_OUTPUT	1091431	0	19326	7133100218	0 / 7133100218	19326 / 1110757
2	POLISH LUCENE MORFOLOGIK FILTER	PRIMARY_OUTPUT	996176	76097	114581	7133024121	76097 / 7133100218	114581 / 1110757
3	HUNSPELL POLISH LUCENE FILTER	PRIMARY_OUTPUT	963133	27267	147624	7133072951	27267 / 7133100218	147624 / 1110757
4	POLISH LUCENE STEMPEL DIRECT	PRIMARY_OUTPUT	791610	43601	319147	7133056617	43601 / 7133100218	319147 / 1110757
5	POLISH LUCENE STEMPEL FILTER	PRIMARY_OUTPUT	791610	43601	319147	7133056617	43601 / 7133100218	319147 / 1110757
6	SNOWBALL POLISH DIRECT	PRIMARY_OUTPUT	718592	70647	392165	7133029571	70647 / 7133100218	392165 / 1110757

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 POLISH LUCENE FILTER	0.000356%	7.234976%
POLISH LUCENE MORFOLOGIK FILTER	0.001024%	2.474799%
Radixor	0.000000%	0.000000%

Oracle-bound pair counts

Stemmer	Unavoidable over errors / gold-negative pairs	Unrepairable under errors / gold-related pairs
HUNSPELL POLISH LUCENE FILTER	25425 / 7133100218	80363 / 1110757
POLISH LUCENE MORFOLOGIK FILTER	73019 / 7133100218	27489 / 1110757
Radixor	0 / 7133100218	0 / 1110757

ALL_CANDIDATES ranking

Rank	Stemmer	Balanced accuracy	Over-stemming (OI)	Under-stemming (UI)
1	Radixor	1.000000	0.000000%	0.000000%
2	POLISH LUCENE MORFOLOGIK FILTER	0.987619	0.001409%	2.474799%
3	HUNSPELL POLISH LUCENE FILTER	0.963822	0.000612%	7.234976%

Classification metrics

Rank	Stemmer	Output policy	Precision	Recall	Specificity	Balanced accuracy	Pairwise accuracy	Error rate
1	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	0.000000
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	0.915099	0.975252	0.999986	0.987619	0.999982	0.000018
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	0.959377	0.927650	0.999994	0.963822	0.999983	0.000017

Pair-relation metrics

Rank	Stemmer	Output policy	F0.5	F1	F2	Jaccard	Fowlkes–Mallows	MCC
1	Radixor	ALL_CANDIDATES	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	0.926529	0.944219	0.962597	0.894332	0.944697	0.944688
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	0.952859	0.943247	0.933827	0.892590	0.943380	0.943372

Raw pair counts

Rank	Stemmer	Output policy	TP	FP	FN	TN	Over error / possible	Under error / possible
1	Radixor	ALL_CANDIDATES	1110757	0	0	7133100218	0 / 7133100218	0 / 1110757
2	POLISH LUCENE MORFOLOGIK FILTER	ALL_CANDIDATES	1083268	100503	27489	7132999715	100503 / 7133100218	27489 / 1110757
3	HUNSPELL POLISH LUCENE FILTER	ALL_CANDIDATES	1030394	43630	80363	7133056588	43630 / 7133100218	80363 / 1110757

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 POLISH LUCENE FILTER	67261	1842	16363	10303	8.625294%	6	130856
POLISH LUCENE MORFOLOGIK FILTER	87092	3078	24406	11666	9.766348%	5	132279
Radixor	19326	0	0	1306	1.093335%	4	120926

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: PL_PL
• 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