Nefnir: A high accuracy lemmatizer for Icelandic

Hrafn Loftsson; J\'on Fri{\dh}rik Da{\dh}ason; Krist\'in Bjarnad\'ottir; Svanhv\'it Lilja Ing\'olfsd\'ottir

arxiv: 1907.11907 · v1 · pith:MU4ZPLBEnew · submitted 2019-07-27 · 💻 cs.CL

Nefnir: A high accuracy lemmatizer for Icelandic

Svanhv\'it Lilja Ing\'olfsd\'ottir , Hrafn Loftsson , J\'on Fri{\dh}rik Da{\dh}ason , Krist\'in Bjarnad\'ottir This is my paper

Pith reviewed 2026-05-24 14:54 UTC · model grok-4.3

classification 💻 cs.CL

keywords Icelandiclemmatizationnatural language processingmorphological databasesuffix substitutionpart-of-speech taggingaccuracy evaluation

0 comments

The pith

Nefnir lemmatizes Icelandic text via suffix substitution rules from a morphological database, reaching 99.55% accuracy on correctly tagged input.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents Nefnir, an open source lemmatizer that converts tagged Icelandic words to their base forms by applying suffix substitution rules extracted from a large morphological database. It reports 99.55% accuracy when input tags are correct and 96.88% when tags come from an automatic part-of-speech tagger. A sympathetic reader would care because lemmatization is a foundational step for many natural language processing tasks in morphologically rich languages, and this approach targets practical accuracy for Icelandic. The evaluation covers both ideal tagging conditions and realistic pipeline use.

Core claim

Nefnir uses suffix substitution rules, derived from a large morphological database, to lemmatize tagged text. Evaluation shows that for correctly tagged text, Nefnir obtains an accuracy of 99.55%, and for text tagged with a PoS tagger, the accuracy obtained is 96.88%.

What carries the argument

Suffix substitution rules derived from a morphological database that replace word endings to produce lemmas based on observed patterns.

Load-bearing premise

The morphological database is comprehensive enough that the suffix substitution rules derived from it will generalize accurately to new Icelandic text outside the database itself.

What would settle it

Running Nefnir on a new Icelandic corpus or set of words absent from the morphological database and measuring whether accuracy falls below 90%.

read the original abstract

Lemmatization, finding the basic morphological form of a word in a corpus, is an important step in many natural language processing tasks when working with morphologically rich languages. We describe and evaluate Nefnir, a new open source lemmatizer for Icelandic. Nefnir uses suffix substitution rules, derived from a large morphological database, to lemmatize tagged text. Evaluation shows that for correctly tagged text, Nefnir obtains an accuracy of 99.55%, and for text tagged with a PoS tagger, the accuracy obtained is 96.88%.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Nefnir is a practical new lemmatizer for Icelandic reporting high accuracies, though evaluation details are limited.

read the letter

Nefnir gives us a new lemmatizer for Icelandic with reported accuracies of 99.55% on gold tags and 96.88% on automatic tags, using suffix rules from a morphological database. This is new in the sense that it is a specific implementation and evaluation for Icelandic, even though the rule-derivation technique is not novel. The paper does well by releasing the tool as open source and providing these concrete performance numbers for a language that needs such tooling. The evaluation is empirical and avoids circularity. That part holds up. The soft spots are in the details. The abstract does not report test set size or error analysis, and there is no discussion of how many test words are outside the database. The stress-test concern applies here: the high numbers might mainly show good coverage in the database rather than robust rule-based generalization. That is a moderate issue for a tool paper, but it does limit how much we can conclude about performance on new text. Readers working on Icelandic computational linguistics or similar languages will find this useful. It is the kind of practical paper that adds a working component to the toolkit. I would recommend sending it to peer review. The contribution is clear enough to warrant referee feedback, even with the gaps in the evaluation description.

Referee Report

2 major / 0 minor

Summary. The paper presents Nefnir, an open-source lemmatizer for Icelandic that derives suffix substitution rules from a large morphological database and applies them to lemmatize tagged text. It reports accuracies of 99.55% when input tags are correct and 96.88% when input comes from an automatic PoS tagger.

Significance. If the reported accuracies reflect genuine generalization to word forms outside the source database, the work would provide a practical, high-accuracy tool for a morphologically rich language together with reproducible open-source code and a simple rule-based method. The empirical framing against an external database is a strength.

major comments (2)

[Evaluation] Evaluation section: the manuscript provides no information on test-set size, the proportion of word forms absent from the morphological database, or any error analysis. Without evidence that a non-trivial fraction of the test material consists of forms unseen in the database, the accuracies of 99.55% and 96.88% cannot be interpreted as measuring generalization via the derived suffix rules rather than database coverage.
[Method] Method section: the description of how suffix-substitution rules are extracted, filtered, and selected from the morphological database is insufficiently detailed to allow reproduction or assessment of whether the rule set is overfit to the database.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate the revisions we will make.

read point-by-point responses

Referee: [Evaluation] Evaluation section: the manuscript provides no information on test-set size, the proportion of word forms absent from the morphological database, or any error analysis. Without evidence that a non-trivial fraction of the test material consists of forms unseen in the database, the accuracies of 99.55% and 96.88% cannot be interpreted as measuring generalization via the derived suffix rules rather than database coverage.

Authors: We agree that these details are necessary for proper interpretation. In the revised manuscript we will report the exact size of the test set, the proportion of word forms absent from the morphological database, and include a short error analysis. This will allow readers to assess the degree of generalization achieved by the suffix rules. revision: yes
Referee: [Method] Method section: the description of how suffix-substitution rules are extracted, filtered, and selected from the morphological database is insufficiently detailed to allow reproduction or assessment of whether the rule set is overfit to the database.

Authors: We acknowledge that the current description is not detailed enough for full reproducibility. We will expand the method section with a step-by-step account of rule extraction, the filtering criteria applied, and the selection procedure used, including any parameters or thresholds. This will also help evaluate potential overfitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes an empirical system that extracts suffix substitution rules from an external morphological database and measures accuracy on held-out tagged text (99.55% gold tags, 96.88% automatic tags). No equations, fitted parameters, or self-citations reduce the reported accuracies to the training inputs by construction; the evaluation is a direct comparison against an independent database on unseen material. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the quality and coverage of an external morphological database for Icelandic; no free parameters are introduced and no new entities are postulated.

axioms (1)

domain assumption Suffix substitution rules derived from a morphological database can accurately map inflected forms to lemmas for Icelandic.
The method assumes Icelandic morphology is sufficiently regular that such rules will cover the majority of cases.

pith-pipeline@v0.9.0 · 5645 in / 1249 out tokens · 53396 ms · 2026-05-24T14:54:18.489742+00:00 · methodology

Nefnir: A high accuracy lemmatizer for Icelandic

Core claim

What carries the argument

Load-bearing premise

What would settle it

discussion (0)