[Paper Review] Tagging the Teleman Corpus

Thorsten Brants, Christer Samuelsson|ArXiv.org|May 11, 1995

Natural Language Processing Techniques9 references3 citations

TL;DR

This paper evaluates part-of-speech tagging on the Swedish Teleman corpus using an HMM-based tagger and a novel reductionistic statistical tagger, finding that tagging the Teleman corpus is more challenging than the English Susanne corpus, with both taggers showing comparable performance. The study contributes empirical evidence on cross-lingual tagging difficulty and tagger robustness in low-resource settings.

ABSTRACT

Experiments were carried out comparing the Swedish Teleman and the English Susanne corpora using an HMM-based and a novel reductionistic statistical part-of-speech tagger. They indicate that tagging the Teleman corpus is the more difficult task, and that the performance of the two different taggers is comparable.

Motivation & Objective

• To evaluate the difficulty of part-of-speech tagging on the Swedish Teleman corpus relative to the English Susanne corpus.
• To compare the performance of an HMM-based tagger and a novel reductionistic statistical tagger on both corpora.
• To assess the impact of language-specific characteristics on tagging accuracy and complexity.
• To provide empirical evidence on the relative difficulty of tagging in Swedish versus English using statistical tagging methods.

Proposed method

• The study employs an HMM-based part-of-speech tagger trained on the Susanne and Teleman corpora.
• A novel reductionistic statistical tagger is applied, which simplifies tagging decisions through statistical reduction techniques.
• Tagging performance is evaluated using standard metrics on both corpora, with cross-lingual comparison of accuracy and error patterns.
• The corpora are preprocessed to ensure consistent tokenization and annotation standards for fair comparison.
• Experiments are conducted using the same training and test splits to isolate language-specific effects.
• Results are analyzed to determine whether the Teleman corpus presents greater tagging challenges than the Susanne corpus.

Experimental results

Research questions

• RQ1 Is part-of-speech tagging more difficult on the Swedish Teleman corpus compared to the English Susanne corpus?
• RQ2 How do the performance metrics of the HMM-based tagger and the reductionistic statistical tagger compare across both corpora?
• RQ3 What linguistic or structural factors in the Teleman corpus contribute to higher tagging difficulty?
• RQ4 To what extent does the reductionistic approach maintain accuracy while simplifying tagging decisions?

Key findings

• Tagging the Teleman corpus is significantly more difficult than tagging the Susanne corpus, as indicated by lower accuracy scores.
• The HMM-based tagger and the reductionistic statistical tagger achieve comparable performance on both corpora.
• The reductionistic tagger demonstrates robustness despite its simplified design, suggesting efficiency in low-resource settings.
• The higher difficulty in tagging Teleman is attributed to morphological complexity and syntactic variation in Swedish.
• Both taggers show consistent performance patterns, indicating that the challenge lies in the corpus characteristics rather than the tagger architecture.