A Spoken Drug Prescription Dataset in French for Spoken Language Understanding

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 89–99 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 89 Turkish Universal Conceptual Cognitive Annotation Necva B¨ol¨uc¨u, Burcu Can Hacettepe University, University of Wolverhampton Department of Computer Engineering, Research Group in Computational Linguistics (RGCL) Ankara Turkey, Wolverhampton UK necvaa@gmail.com, b.can@wlv.ac.uk Abstract Universal Conceptual Cognitive Annotation (UCCA) is a cross-lingual semantic annotation framework that provides an easy annotation without any requirement for linguistic background. UCCA-annotated datasets have been already released in English, French, and German. In this paper, we introduce the first UCCA-annotated Turkish dataset that currently involves 50 sentences obtained from the METU-Sabanci Turkish Treebank. We followed a semi-automatic annotation approach, where an external semantic parser is utilised for an initial annotation of the dataset, which is partially accurate and requires refinement. We manually revised the annotations obtained from the semantic parser that are not in line with the UCCA rules that we defined for Turkish. We used the same external semantic parser for evaluation purposes and conducted experiments with both zero-shot and few-shot learning. This is the initial version of the annotated dataset and we are currently extending the dataset. We are releasing the current Turkish UCCA annotation guideline along with the annotated dataset. Keywords: UCCA, semantic parsing, semantic annotation, Turkish, low-resource languages 1. Introduction Semantics is concerned with anything related to mean- ing and explores answers to such questions (Vigliocco and Vinson, 2007): 1) How do we represent the mean- ing of each word? 2) How do the meanings of words form a semantic structure? 3) How are words related to each other in a semantic structure? Word embed- dings allow the meaning of a word to be represented in a compact low-dimensional space. It is a learned rep- resentation where words with a similar meaning have similar representations. The first breakthrough in neu- ral word representation is word2vec (Mikolov et al., 2013), a statistical-based method (built on a shallow neural architecture) for learning word embeddings via their contextual information obtained from a prefer- ably large corpus. Learning high-quality word embed- dings using neural methods rather than statistical meth- ods such as co-occurrence matrices has improved the performance of all natural language processing (NLP) tasks such as text classification (Wensen et al., 2016; Li et al., 2018) and machine translation (Jansen, 2017; Qi et al., 2018), which applies to all languages with available large resources. Contextualised word embed- dings such as BERT (Devlin et al., 2018) and its varia- tions (Sanh et al., 2019; Yang et al., 2019) have recently been introduced, and have shown even better perfor- mance in almost any NLP task. Better word representation techniques have paved the way for semantic representation for compositional se- mantics for bigger text units, such as phrases and sen- tences. Semantic representation is the process of map- ping a given text into a formal or abstract form that can be understood by machines. Semantic representa- tion is widely dominated by graph …

Content truncated to 500 words