GermDetect - Verb Placement Error Detection Datasets for Learners of Germanic Languages - Abstract

Sep 14, 2026

Uploaded by Logic

GermDetect provides Universal Dependencies (UD)-based, linguistically informed verb placement error detection datasets for learners of Germanic languages, framed as a token-level classification task. It addresses the underrepresentation of naturalistic word order and verb placement errors in existing benchmark resources for grammatical error detection/correction (GED/GEC) and linguistic acceptability assessment (LA). Multilingual mBERT models trained on GermDetect demonstrate improved detection when errors are induced using UD-informed methods, with ablation results showing gains for lower-resource languages via inclusion of structurally related languages.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Abstract

1 Introduction

2 Related Work

2.1 Verb Placement Rules in Germanic Languages

GermDetect Verb Placement Error Detection Datasets for Learners of Germanic Languages Noah-Manuel Michael Kiel University, Germany Leibniz Institute for Science and Mathematics Education, Kiel, Germany Linköping University, Sweden [michael@ipn.uni-kiel.de](michael@ipn.uni-kiel.de) Andrea Horbach Kiel University, Germany Leibniz Institute for Science and Mathematics Education, Kiel, Germany [horbach@ipn.uni-kiel.de](horbach@ipn.uni-kiel.de) Abstract Correct verb placement is difficult to acquire for second-language (L2) learners of Germanic languages. However, word order errors and, consequently, verb placement errors, are heavily underrepresented in benchmark datasets of NLP tasks such as grammatical error detection (GED)/correction (GEC) and linguistic acceptability assessment (LA) . If they are present, they are most often naively introduced, or classification occurs at the sentence level, preventing the precise identification of individual errors and the provision of appropriate feedback to learners. To remedy this, we present GermDetect: Universal Dependencies-based (UD), linguistically informed verb placement error detection datasets for learners of Germanic languages, designed as a token classification task. As our datasets are UD-based, we are able to provide them in most major Germanic languages: Afrikaans, German, Dutch, Faroese, Icelandic, Danish, Norwegian (Bokmål and Nynorsk), and Swedish. We train multilingual BERT (mBERT) models on GermDetect and show that linguistically informed, UD-based error induction results in more effective models for verb placement error detection than models trained on naively introduced errors. Finally, we conduct ablation studies on multilingual training and find that lower-resource languages benefit from the inclusion of structurally related languages in training. 1 Introduction Correct verb placement is difficult to acquire for L2 learners of Germanic languages. This is due to the placement depending on different factors such as finiteness and the clause type in which the verbs occur. Example (1) illustrates a Dutch sentence consisting of a single main clause. (1) Hij heeft een hond gekocht. he has a dog bought S V2 O O V ‘He has bought a dog.’ Two characteristics can be observed here: The finite verb heeft is placed in the second position (V2) and the non-finite, participle verb gekocht is placed clause-finally. Example (2), consisting of a main and a subordinate clause, illustrates how the verb placement changes: Here, the finite weet occupies the V2 position in the main clause, but both the finite heeft and non-finite gekocht are placed clause-finally in the subordinate clause, following subject-object-verb (SOV) word order. (2) Ik weet dat hij een hond heeft gekocht. I know that he a dog has bought S V2 _ S O O V V ‘I know that he has bought a dog.’ These and other context-dependent changes in verb placement in Germanic languages are not trivial for L2 learners. Therefore, being able to provide accurate feedback to learners about their verb placement is crucial. However, word order errors are heavily underrepresented in both GED/GEC shared task datasets and LA datasets. If they are present, they are often introduced naively, which means they do not represent the kinds of errors L2 learners are likely to make. This leaves the capability of GED systems to detect naturalistic word order errors underexplored. Verb placement errors, as a subset of word order errors, are even more poorly represented, which, in the context of Germanic languages, is particularly critical. Additionally, LA datasets are often formulated as sentence-level binary classification or pair-wise ranking tasks. This does not allow for locating errors within a sentence and therefore offers 818 Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications, pages 818–829 July 31-August 1, 2025 ©2025 Association for Computational Linguistics

FAQs

What problem does GermDetect target for learners of Germanic languages?

GermDetect targets the difficulty L2 learners face with context-dependent verb placement, and the lack of naturalistic verb placement errors in common GED/GEC and LA benchmark datasets.

How is GermDetect designed for error detection?

GermDetect is built from Universal Dependencies and uses a token classification setup so individual verb placement errors can be located within a sentence.

Which languages are covered by the GermDetect datasets?

The datasets are available for major Germanic languages including Afrikaans, German, Dutch, Faroese, Icelandic, Danish, Norwegian (Bokmål and Nynorsk), and Swedish.

Sep 14, 2026

Uploaded by Logic