---
title: "Multilingual Dependency Analysis with a — McDonald 2006"
description: "The paper addresses labeled dependency parsing for many languages at once, using the CoNLL-X shared task treebanks for 13 typologically varied languages."
authors: ["Ryan McDonald", "Kevin Lerman", "Fernando Pereira"]
year: 2006
venue: "Proceedings of the Tenth Conference on Computational Natural Language Learning (CoNLL-X)"
source: "https://aclanthology.org/W06-2932/"
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/papers/mcdonald-2006-multilingual-dependency-analysis-two-stage/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/mcdonald-2006-multilingual-dependency-analysis-two-stage/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2006](https://md.research.elamur.ai/free/papers/2006/) / Multilingual Dependency Analysis with a Two-Stage Discriminative Parser

# Multilingual Dependency Analysis with a Two-Stage Discriminative Parser

**Authors:** [Ryan McDonald](https://md.research.elamur.ai/free/authors/ryan-mcdonald/), Kevin Lerman, [Fernando Pereira](https://md.research.elamur.ai/free/authors/fernando-pereira/) · **Year:** 2006 · **Venue:** Proceedings of the Tenth Conference on Computational Natural Language Learning (CoNLL-X) · **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/W06-2932/) · [PDF](https://aclanthology.org/W06-2932.pdf) · [OpenAlex](https://openalex.org/W2143541936)

**Citations:** 219 citations (openalex-snapshot · 2026-10-14) · [102](#cited-by) in our library, of the 140673 papers whose bibliographies we read

## Summary
The paper addresses labeled dependency parsing for many languages at once, using the CoNLL-X shared task treebanks for 13 typologically varied languages. The authors split the task into two parts. First, an unlabeled parser from the maximum spanning tree family (McDonald and Pereira, 2006), trained with the online large-margin learner MIRA, finds head–dependent pairs. It can run in an exact projective mode or an approximate non-projective mode. Where the data provided morphology, the parser was extended with morphological attributes of the head, the dependent, and their pairwise combinations. Second, a separate labeler assigns syntactic categories. It treats all dependents of one head as a sequence, scores it with a first-order Markov model decoded by Viterbi, and also trains with MIRA. Its features can look at the whole fixed tree. Averaged over the 13 languages, the system reaches 87.0 unlabeled and 80.8 labeled accuracy and beats the shared-task mean for every language. This shows that one discriminative design transfers across languages when an annotated treebank exists. The error analysis identifies verbs, conjunctions and prepositions as the weakest points.

## Key points
- Test results (UA/LA) range from 92.8/90.7 for Japanese down to 74.7/63.2 for Turkish and 79.3/66.9 for Arabic. Slovene, Arabic and Turkish stay below 80 unlabeled accuracy, which the authors link to small training sets and/or rich inflection with loose word order.
- Projective or non-projective decoding, and lemmas or inflected forms (lemmas chosen for Arabic and Spanish), were selected per language on held-out data. Using the non-projective parser everywhere would not change results significantly.
- An ablation over the 10 smallest datasets shows the full system at 86.3/79.7 UA/LA versus 84.8/77.7 when it is forced projective, uses atomic labeling and has no morphology. Each component alone gives only a small gain.
- Non-projective parsing lifted Dutch from 78.8/74.7 to 83.6/79.2. Morphological features added 1–3% absolute for inflected languages such as Spanish, Arabic, Turkish and Slovene. Sequential labeling mainly helped separate subjects from objects, for example raising Swedish object precision/recall from 81.7/75.6 to 84.2/81.3.
- For Arabic, the parser and the labeler are each about 79% accurate, but their errors rarely coincide, so labeled accuracy drops to 67%. Training on 200, 400 and 800 sentences and on the full set gives 54, 60, 62 and 67 LA.

## Interesting ideas
- **Use a pipeline to unlock global features.** Once the tree is fixed, the labeler can use cues that a tractable joint parser cannot, such as child counts, sibling properties, grandparent morphology and whether an edge crosses others. This pattern suits any structured task where exact joint inference limits the features you can use.
- **Label sibling edges jointly.** Dependents of the same head tend to have correlated roles; knowing the subject helps find the object. Treating them as a short sequence for Viterbi decoding is a cheap way to capture such interactions.
- **Encode agreement with cross-product morphology features.** Adding head attributes, dependent attributes and all their pairings lets the model learn gender, case and number agreement directly. This could be valuable for other highly inflected languages.
- **Report each stage's accuracy separately.** Measuring labeling accuracy independently of head correctness revealed that the two Arabic stages fail on different word classes. That explains the large combined drop, which a single end-to-end score would hide.
- **Ablate component combinations, not just single components.** Individual removals looked insignificant, but removing everything together caused a clear loss. Testing combinations gives a more honest view of what a system's design contributes.

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

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/)

Topics: [Treebanks](https://md.research.elamur.ai/free/topics/treebanks/)

## Cited by

Cited by

102 papers of our library cite it (of the 140673 whose bibliographies we read); 12 of them have a page here, 90 not yet.

- [Evaluation of Dependency Parsers on Unbounded Dependencies](https://md.research.elamur.ai/free/papers/aclbulk-2010-nivre-evaluation-dependency-parsers-unbounded-dependencies/)
- [Turning on the Turbo: Fast Third-Order Non-Projective Turbo Parsers](https://md.research.elamur.ai/free/papers/aclbulk-2013-martins-turning-turbo-fast-third-order/)
- [Correcting Dependency Annotation Errors](https://md.research.elamur.ai/free/papers/dickinson-2009-correcting-dependency-annotation-errors/)
- [Latent Structure Perceptron with Feature Induction for Unrestricted Coreference Resolution](https://md.research.elamur.ai/free/papers/fernandes-2012-latent-structure-perceptron-feature-induction/)
- [Statistical Parsing by Machine Learning from a Classical Arabic Treebank](https://md.research.elamur.ai/free/papers/langstream-2015-dukes-statistical-parsing-machine-learning-classical/)
- [Turbo Parsers: Dependency Parsing by Approximate Variational Inference](https://md.research.elamur.ai/free/papers/martins-2010-turbo-parsers-dependency-parsing-approximate/)
- [Characterizing the Errors of Data-Driven Dependency Parsing Models](https://md.research.elamur.ai/free/papers/missing4-2007-mcdonald-characterizing-errors-data-driven-dependency/)
- [Algorithms for Deterministic Incremental Dependency Parsing](https://md.research.elamur.ai/free/papers/narrative-2008-nivre-algorithms-deterministic-incremental-dependency-parsing/)
- [Non-Projective Dependency Parsing in Expected Linear Time](https://md.research.elamur.ai/free/papers/nivre-2009-non-projective-dependency-parsing-expected/)
- [CoNLL-X Shared Task on Multilingual Dependency Parsing](https://md.research.elamur.ai/free/papers/parsers-2006-buchholz-conll-x-shared-task-multilingual/)
- [Does Universal Dependencies need a parsing representation? An investigation of English](https://md.research.elamur.ai/free/papers/silveira-2015-does-universal-dependencies-need-parsing/)
- [Semantic Dependency Parsing of NomBank and PropBank: An Efficient Integrated Approach via a Large-scale Feature Selection](https://md.research.elamur.ai/free/papers/zhao-2009-semantic-dependency-parsing-nombank-propbank/)

## Cite

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

```bibtex
@inproceedings{mcdonald2006multilingual,
  title = {{Multilingual Dependency Analysis with a Two-Stage Discriminative Parser}},
  author = {McDonald, Ryan and Lerman, Kevin and Pereira, Fernando},
  booktitle = {Proceedings of the Tenth Conference on Computational Natural Language Learning (CoNLL-X)},
  year = {2006},
  doi = {10.3115/1596276.1596317},
  url = {https://doi.org/10.3115/1596276.1596317}
}
```

- [← previous Morpho-syntactic Information for Automatic Error Analysis of Statistical Machine Translation Output](https://md.research.elamur.ai/free/papers/popovic-2006-morpho-syntactic-information-automatic-error/)
- [next → Noun Phrase Chunking in Hebrew: Influence of Lexical and Morphological Features](https://md.research.elamur.ai/free/papers/aclbulk-2006-goldberg-noun-phrase-chunking-hebrew-influence/)
- [↑ back to papers of 2006](https://md.research.elamur.ai/free/papers/2006/)

## References

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

1. S. Buchholz, E. Marsi, A. Dubey, and Y. Krymolowski. 2006. CoNLL-X shared task on multilingual dependency parsing. SIGNLL. [find](https://scholar.google.com/scholar?q=S%2E+Buchholz%2C+E%2E+Marsi%2C+A%2E+Dubey%2C+and+Y%2E+Krymolowski%2E+2006%2E+CoNLL%2DX+shared+task+on+multilingual+dependency+parsing%2E+SIGNLL%2E) [our card](https://md.research.elamur.ai/free/papers/parsers-2006-buchholz-conll-x-shared-task-multilingual/)
2. K. Crammer and Y. Singer. 2003. Ultraconservative online algorithms for multiclass problems. JMLR. [find](https://scholar.google.com/scholar?q=K%2E+Crammer+and+Y%2E+Singer%2E+2003%2E+Ultraconservative+online+algorithms+for+multiclass+problems%2E+JMLR%2E)
3. Y. Ding and M. Palmer. 2005. Machine translation using probabilistic synchronous dependency insertion grammars. In Proc. ACL. [find](https://scholar.google.com/scholar?q=Y%2E+Ding+and+M%2E+Palmer%2E+2005%2E+Machine+translation+using+probabilistic+synchronous+dependency+insertion+grammars%2E+In+Proc%2E+ACL%2E) [our card](https://md.research.elamur.ai/free/papers/ding-2005-machine-translation-probabilistic-synchronous-dependency/)
4. A. Haghighi, A. Ng, and C. Manning. 2005. Robust textual inference via graph matching. In Proc. HTL-EMNLP. [find](https://scholar.google.com/scholar?q=A%2E+Haghighi%2C+A%2E+Ng%2C+and+C%2E+Manning%2E+2005%2E+Robust+textual+inference+via+graph+matching%2E+In+Proc%2E+HTL%2DEMNLP%2E)
5. R. Hudson. 1984. Word Grammar. Blackwell. [find](https://scholar.google.com/scholar?q=R%2E+Hudson%2E+1984%2E+Word+Grammar%2E+Blackwell%2E)
6. R. McDonald and F. Pereira. 2006. Online learning of approximate dependency parsing algorithms. In Proc. EACL. [find](https://scholar.google.com/scholar?q=R%2E+McDonald+and+F%2E+Pereira%2E+2006%2E+Online+learning+of+approximate+dependency+parsing+algorithms%2E+In+Proc%2E+EACL%2E) [our card](https://md.research.elamur.ai/free/papers/continual-2006-mcdonald-online-learning-approximate-dependency-parsing/)
7. R. McDonald, K. Crammer, and F. Pereira. 2005a. Online large-margin training of dependency parsers. In Proc. ACL. [find](https://scholar.google.com/scholar?q=R%2E+McDonald%2C+K%2E+Crammer%2C+and+F%2E+Pereira%2E+2005a%2E+Online+large%2Dmargin+training+of+dependency+parsers%2E+In+Proc%2E+ACL%2E) [our card](https://md.research.elamur.ai/free/papers/narrative-2005-mcdonald-online-large-margin-training-dependency/)
8. R. McDonald, F. Pereira, K. Ribarov, and J. Hajič. 2005b. Non-projective dependency parsing using spanning tree algorithms. In Proc. HLT-EMNLP. [find](https://scholar.google.com/scholar?q=R%2E+McDonald%2C+F%2E+Pereira%2C+K%2E+Ribarov%2C+and+J%2E+Hajic%CC%8C%2E+2005b%2E+Non%2Dprojective+dependency+parsing+using+spanning+tree+algorithms%2E+In+Proc%2E+HLT%2DEMNLP%2E) [our card](https://md.research.elamur.ai/free/papers/narrative-2005-mcdonald-non-projective-dependency-parsing-spanning/)
9. R. McDonald. 2006. Discriminative sentence compression with soft syntactic constraints. In Proc. EACL. [find](https://scholar.google.com/scholar?q=R%2E+McDonald%2E+2006%2E+Discriminative+sentence+compression+with+soft+syntactic+constraints%2E+In+Proc%2E+EACL%2E)
10. I.A. Mel 0 čuk. 1988. Dependency Syntax: Theory and Practice. State University of New York Press. [find](https://scholar.google.com/scholar?q=I%2EA%2E+Mel+0+c%CC%8Cuk%2E+1988%2E+Dependency+Syntax%3A+Theory+and+Practice%2E+State+University+of+New+York+Press%2E)
11. J. Nivre. 2005. Dependency grammar and dependency parsing. Technical Report MSI report 05133, Växjö University: School of Mathematics and Systems Engineering. [find](https://scholar.google.com/scholar?q=J%2E+Nivre%2E+2005%2E+Dependency+grammar+and+dependency+parsing%2E+Technical+Report+MSI+report+05133%2C+Va%CC%88xjo%CC%88+University%3A+School+of+Mathematics+and+Systems+Engineering%2E)