---
title: "Tagging English by Path Voting Constraints — Tür 1998"
description: "Taggers built from hand-written constraints usually delete a reading as soon as a rule… 97.94 precision. 15 references, cited by 3 papers here."
authors: ["Gökhan Tür", "Kemal Oflazer"]
year: 1998
venue: "36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Volume 2"
source: "https://aclanthology.org/P98-2208/"
doi: "10.3115/980691.980777"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/tur-1998-tagging-english-path-voting-constraints/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/tur-1998-tagging-english-path-voting-constraints/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [1998](https://md.research.elamur.ai/free/en/papers/1998/) / Tagging English by Path Voting Constraints

# Tagging English by Path Voting Constraints

**Authors:** [Gökhan Tür](https://md.research.elamur.ai/free/en/authors/gokhan-tur/) and [Kemal Oflazer](https://md.research.elamur.ai/free/en/authors/kemal-oflazer/) · **Year:** 1998 · **Venue:** [36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Volume 2](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 1998](https://md.research.elamur.ai/free/en/venues/acl/1998/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/P98-2208/) · [PDF](https://aclanthology.org/P98-2208.pdf) · [DOI 10.3115/980691.980777](https://doi.org/10.3115/980691.980777) · [OpenAlex](https://openalex.org/W2125622217)

**Citations:** 6 citations (OpenAlex, 2026-10-14) · 3 in our library, of the 149,439 papers whose bibliographies we read

## Summary
Taggers built from hand-written constraints usually delete a reading as soon as a rule fires. That makes the order of rules matter, and conflicting rules become hard to manage. Tür and Oflazer carry their voting-constraint scheme, first built for Turkish morphological disambiguation, over to English.

Each sentence becomes a lattice holding every candidate tag of every token. Each rule is a short pattern over tags and/or word forms, paired with a positive or negative integer vote. Any lattice path that contains a match collects that vote. The highest-scoring path is chosen only after all rules have spoken, so the result does not depend on rule order. A sliding-window search, similar to Viterbi, keeps tagging fast.

The rules come from two sources:
- 400 tag bigrams and trigrams scored on the training data;
- 824 manual rules, mostly negative or lexicalized.

The full system reaches 97.50% test accuracy on WSJ and 96.67% on Brown, both from the Penn Treebank and under a closed vocabulary. This is comparable to stochastic taggers and Brill's tagger. The system can also leave controlled ambiguity in its output.

## Key points
- Evaluation setup:
  - WSJ: a 5500-sentence portion with 11-fold cross-validation (about 118,500 training words and 11,800 test words per fold).
  - Brown: 4000 sentences (about 100,000 words) with 10-fold cross-validation.
- Accuracy by rule set:
  - Lexical votes alone: 94.54% test (WSJ).
  - Adding 200 bigrams, 200 trigrams and 824 manual rules: 97.89% train and 97.50% test (WSJ); 97.18% and 96.67% (Brown).
- Votes for statistical k-grams:
  - p is a smoothed ratio of how often a tag sequence actually occurs to how often it could occur, computed as (f+0.5)/(n+1).
  - The vote is p minus its standard error, times 100.
  - Each path also starts with lexical tag-frequency votes, scaled to 0–100.
- The manual rules were written from errors on one WSJ training fold and reused unchanged on Brown.
- Speed and ambiguity:
  - The tagger runs at about 1600 tokens/sec with over 1000 rules loaded (the longest spans 5 tokens).
  - Keeping partial paths within a ratio ρ of the best one trades precision for recall. At ρ=0.99: recall 97.94, precision 96.70, ambiguity 1.012. At ρ=0.91: recall 99.13, precision 86.68, ambiguity 1.143.

## Interesting ideas
- Rules never delete readings directly; they only add or subtract score on lattice paths, and the decision comes at the end, so the rule set can grow without ordering conflicts.
- A rule's strength is its observed success rate minus its standard error, so rare patterns with high ratios do not outrank well-attested ones.
- Learned and manual rules share one vote scale, and manual votes are set large enough to override statistics; the manual rules were written by inspecting where the statistical rules failed.
- Keeping all paths within a fixed ratio of the best score turns a single-best tagger into a tunable ambiguous one, reported with recall, precision and average ambiguity.
- Manual rules developed on WSJ were applied unchanged to Brown, which gives a cheap check of how well hand rules transfer across genres.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tokens): The tagger in crate en could express its disambiguation rules as additive votes over a tag lattice, which keeps each decision explainable as a list of the rules that fired.
- **I2** (learn): Self-induced repair rules could be ranked by this kind of confidence lower bound before they are accepted.
- **I3** (learn): Hand-written fixes could be layered over silver-data-derived rules on the same scale, each targeting an observed failure.
- **I4** (tree): The parsers could receive a small set of near-best tags on uncertain tokens instead of one forced choice, with the ambiguity level measured explicitly.
- **I5** (learn): Rules induced on one treebank or genre could be evaluated on a held-out genre as a standing negative control against overfitting.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Bilkent University](https://md.research.elamur.ai/free/en/universities/bilkent-university/) — University, Ankara, Türkiye · 5 works · 490 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works

## Cite

Cite as: Gökhan Tür and Kemal Oflazer. 1998. Tagging English by Path Voting Constraints. 36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Volume 2.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{tur1998tagging,
  title = {{Tagging English by Path Voting Constraints}},
  author = {Gökhan Tür and Kemal Oflazer},
  booktitle = {36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Volume 2},
  year = {1998},
  doi = {10.3115/980691.980777},
  url = {https://doi.org/10.3115/980691.980777}
}
```

- [← previous Generating Determiners and Quantifiers in Hebrew](https://md.research.elamur.ai/free/en/papers/netzer-1998-generating-determiners-quantifiers-hebrew/)
- [next → Syntactic and Semantic Transfer with F-Structures](https://md.research.elamur.ai/free/en/papers/dorna-1998-syntactic-semantic-transfer-f-structures/)
- [↑ back to papers of 1998](https://md.research.elamur.ai/free/en/papers/1998/)

## References

15 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Eric Brill. 1992. [A simple-rule based part-of-speech tagger](https://md.research.elamur.ai/free/en/papers/brill-1992-simple-rule-based-part-speech/). In Proceedings of the Third Conference on Applied Natural Language Processing, Trento, Italy.
2. Eric Brill. 1994. Some advances in rule-based part of speech tagging. In Proceedings of the Twelfth National Conference on Articial Intelligence (AAAI-94), Seattle, Washinton. [find](https://scholar.google.com/scholar?q=Eric+Brill%2E+1994%2E+Some+advances+in+rule%2Dbased+part+of+speech+tagging%2E+In+Proceedings+of+the+Twelfth+National+Conference+on+Articial+Intelligence+%28AAAI%2D94%29%2C+Seattle%2C+Washinton%2E)
3. Eric Brill. 1995. [Transformation-based error-driven learning and natural language processing: A case study in part-of-speech tagging](https://md.research.elamur.ai/free/en/papers/brill-1995-transformation-based-error-driven-learning/). Computational Linguistics, 21(4):543-566, December.
4. Kenneth W. Church. 1988. A stochastic parts program and a noun phrase parser for unrestricted text. In Proceedings of the Second Conference on Applied Natural Language Processing, Austin, Texas. [find](https://scholar.google.com/scholar?q=Kenneth+W%2E+Church%2E+1988%2E+A+stochastic+parts+program+and+a+noun+phrase+parser+for+unrestricted+text%2E+In+Proceedings+of+the+Second+Conference+on+Applied+Natural+Language+Processing%2C+Austin%2C+Texas%2E)
5. Doug Cutting, Julian Kupiec, Jan Pedersen, and Penelope Sibun. 1992. [A practical part-of-speech tagger](https://md.research.elamur.ai/free/en/papers/cutting-1992-practical-part-speech-tagger/). In Proceedings of the Third Conference on Applied Natural Language Processing, Trento, Italy.
6. Steven J. DeRose. 1988. [Grammatical category disambiguation by statistical optimization](https://md.research.elamur.ai/free/en/papers/derose-1988-grammatical-category-disambiguation-statistical-optimization/). Computational Linguistics, 14(1):31-39. [↗](https://doi.org/10.5555/49084.49087)
7. Fred Karlsson, Atro Voutilainen, Juha tteikkil~, and Arto Anttila. 1995. Constraint Grammar-A Language-Independent System for Parsing Unrestricted Text. • Mouton de Gruyter. [find](https://scholar.google.com/scholar?q=Fred+Karlsson%2C+Atro+Voutilainen%2C+Juha+tteikkil%7E%2C+and+Arto+Anttila%2E+1995%2E+Constraint+Grammar%2DA+Language%2DIndependent+System+for+Parsing+Unrestricted+Text%2E+%E2%80%A2+Mouton+de+Gruyter%2E)
8. Kemal Oflazer and GSkhan Tiir. 1997. Morphological disambiguation by voting constraints. In Proceedings of ACL '97/EACL '97, The 35th Annual Meeting of the Association for Computational Linguistics, June. [↗](https://doi.org/10.48550/arxiv.cmp-lg/9704011)
9. Beatrice Santorini. 1995. Part-of-speech tagging guidelines fro the penn treebank project. Available at h t t p ://www. l d c . u p e n n , edu/. 3rd Revision, 2rid Printing. [find](https://scholar.google.com/scholar?q=Beatrice+Santorini%2E+1995%2E+Part%2Dof%2Dspeech+tagging+guidelines+fro+the+penn+treebank+project%2E+Available+at+h+t+t+p+%3A%2F%2Fwww%2E+l+d+c+%2E+u+p+e+n+n+%2C+edu%2F%2E+3rd+Revision%2C+2rid+Printing%2E)
10. Evelyne Tzoukermann, Dragomir R. Radev, and William A. Gale. 1995. Combining linguistic knowledge and statistical learning in french part-of-speech tagging. In Proceedings of the ACL SIGDAT Workshop From Texts to Tags: Issues in Multilingual Language Analysis, pages 51-57. [find](https://scholar.google.com/scholar?q=Evelyne+Tzoukermann%2C+Dragomir+R%2E+Radev%2C+and+William+A%2E+Gale%2E+1995%2E+Combining+linguistic+knowledge+and+statistical+learning+in+french+part%2Dof%2Dspeech+tagging%2E+In+Proceedings+of+the+ACL+SIGDAT+Workshop+F)
11. Atro Voutilainen and Pasi Tapanainen. 1993. Ambiguity resolution in a reductionistic parser. In Proceedings of EACL'93, Utrecht, Holland. [find](https://scholar.google.com/scholar?q=Atro+Voutilainen+and+Pasi+Tapanainen%2E+1993%2E+Ambiguity+resolution+in+a+reductionistic+parser%2E+In+Proceedings+of+EACL%2793%2C+Utrecht%2C+Holland%2E)
12. Atro Voutilainen, Juha Heikkil~, and Arto Anttila. 1992. Constraint Grammar of English. University of Helsinki. [find](https://scholar.google.com/scholar?q=Atro+Voutilainen%2C+Juha+Heikkil%7E%2C+and+Arto+Anttila%2E+1992%2E+Constraint+Grammar+of+English%2E+University+of+Helsinki%2E)
13. Atro Voutilainen. 1995a. Morphological disambiguation. In Fred Karlsson, Atro Voutilainen, Juha Heikkil~, and Arto Anttila, editors, Constraint Grammar-A Language-Independent System for Parsing Unrestricted Text, chapter 5. Mouton de Gruyter. [find](https://scholar.google.com/scholar?q=Atro+Voutilainen%2E+1995a%2E+Morphological+disambiguation%2E+In+Fred+Karlsson%2C+Atro+Voutilainen%2C+Juha+Heikkil%7E%2C+and+Arto+Anttila%2C+editors%2C+Constraint+Grammar%2DA+Language%2DIndependent+System+for+Parsing+Unrest)
14. Atro Voutilainen. 1995b. A syntax-based part-of-speech analyzer. In Proceedings of the Seventh Conference of the European Chapter of the Association of Computational Linguistics, Dublin, Ireland. [find](https://scholar.google.com/scholar?q=Atro+Voutilainen%2E+1995b%2E+A+syntax%2Dbased+part%2Dof%2Dspeech+analyzer%2E+In+Proceedings+of+the+Seventh+Conference+of+the+European+Chapter+of+the+Association+of+Computational+Linguistics%2C+Dublin%2C+Ireland%2E)
15. Ralph Weischedel, Marie Meteer, Richard Schwartz, Lance Ramshaw, and Jeff Pal-mucci. 1993. [Coping with ambiguity and unknown words through probabilistic models](https://md.research.elamur.ai/free/en/papers/weischedel-1993-coping-ambiguity-unknown-words-through/). Computational Linguistics, 19(2):359-382.

*Updated 1 November 2026*