Publications using this dataset must include a reference to the following publication:
Ekaterina Kochmar, Sian Gooding, Matthew Shardlow (2020). Detecting Multiword Expression Type Helps Lexical Complexity Assessment. In Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4428–4437.
The full annotated dataset consisting of 4732 phrases extracted from the Complex word identification Shared Task 2018 dataset is stored in the tab-separated file final_MWE_dataset.tsv with the following fields:
- Unique identifier of the expression in this dataset (
Index): from 0 to 4731 - Annotator's id (
Annotator): 0 for the cases that were annotated by all three annotators, and 1-3 for individual annotations from each of the annotators - The original subset in the shared task data the instance comes from (
Subset): Train, Dev or Test - Genre of the text the instance comes from (
Genre): News, Wikipedia or WikiNews - The unique identifier of the text in the original shared task data (
ID) - Binary compexity score from the original shared task data (
Complex_binary): 0 for non-complex and 1 for complex expressions - Complexity score assigned based on the proportion of 20 annotators that considered an expression complex; derived from the original dataset (
Complex_probabilistic): values in the range of [0.0, 1.0] - Character offset for the beginning of the expression in the instance; derived from the original dataset (
Start_index) - Character offset for the end of the expression in the instance; derived from the original dataset (
End_index) - Number of native annotators, among the total of 10, that considered an expression complex; derived from the original dataset (
Native_complex): values in the range of 0 to 10 - Number of non-native annotators, among the total of 10, that considered an expression complex; derived from the original dataset (
Non_native_complex): values in the range of 0 to 10 - The original expression from the shared task dataset that was annotated with an MWE type (
Expression) - The original context of use in the shared task dataset (
Context) - MWE type assgined to the expression (
Annotation) – see the paper for details on the annotation scheme.
Annotation guidelines used in this project are available here.
This work is licensed under a Creative Commons Attribution-NonCommerial-ShareAlike 4.0 International License.
