jonathanmutal/Medical-Questionnaire-Multilingual-Translation

Diagnostic Interview Corpus

The Diagnostic interview corpus is a multilingual dataset of 12,754 French medical diagnostic interview sentences (questions and instructions) with translations into 12 target languages and a semantic gloss based on UMLS (Unified Medical Language System).

It supports research on:

• Low-resource multilingual medical machine translation
• Semantic representation based on UMLS concepts
• Generation of pictograph sequences based on concept sequences for patients with limited health literacy.

To access concept–pictograph pairs, please refer to this repository.

---

Languages

• Source: French
• Targets: The following table lists the target languages together with their codes as used in the datasets. The codes follow ISO 639-1 when available (e.g., sq for Albanian, es for Spanish), and ISO 639-3 or extended conventions when needed (e.g., prs for Dari, ary for Moroccan Arabic, en-simple for Simple English, and umls for semantic glosses).

Language	Code
Albanian	sq
Modern Standard Arabic	ar
Moroccan Arabic	ary
Tunisian Arabic	aeb
Algerian Arabic (Dziria)	arq
Dari	prs
Farsi	fa
Russian	ru
Simple English	en-simple
Spanish	es
Tigrinya	ti
Ukrainian	uk
UMLS (semantic glosses)	umls

• Semantic gloss: Representation as a sequence of concepts, using UMLS concepts for medical concepts and custom concepts for functional elements such as agents or modes. Two versions of the gloss are provided: 1) using concept names, 2) using UMLS CUIs (Concept Unique Identifiers) for UMLS concepts and names for custom concepts. Note: as opposed to the CUIs, some UMLS concept names may change over time due to UMLS updates.

---

Dataset

The dataset includes one file:

• translations.csv: French sentences with their human translations into the target languages.

translation.csv

Data Structure

Each row in the table corresponds to a target language (tgt_lang) and, where applicable, a gender variant (default / female).

• sentence_id: Unique identifier for the source sentence (shared across languages/variants)
• src_lang: Source language code
• tgt_lang: Target language code
• gender_variant: Some target languages in the corpus use grammatical gender in ways that affect medical communication. For these cases, the dataset includes two translation variants:
  • default: gender-neutral or conventionally masculine (e.g., Albanian jeni student? “Are you a student?”)
  • female: explicitly marked female form (e.g., Albanian jeni studente?)
• source_text: Original French sentence
• target_text: Translation into the target language
• semantic_gloss: Semantic representation: pipe-separated sequence of concepts using concept names (UMLS + custom concepts).
• CUI_semantic_gloss: Semantic representation: pipe-separated sequence of concepts using CUIs for UMLS concepts and names for custom concepts (aligned 1:1 with semantic_gloss)

Distribution

• 12,754 unique French sentences
• 12 parallel translations
• Two semantic gloss representations per sentence: one using concept names, and one using CUIs for UMLS concepts

The following table summarises the data by language. Each row corresponds to a target language (tgt_lang) and, where applicable, a gender variant (default / female).

• tgt_lang: Target language code
• gender_variant: default or female
• n_rows: total rows for this language/variant (from both files)
• n_unique_sentences = unique sentences for this language/variant
• total_rows = overall rows across all gender variants for this language
• total_unique_sentences = overall unique sentences across all gender variants for this language

tgt_lang	gender_variant	n_rows	n_unique_sentences	total_rows	total_unique_sentences
aeb	default	11084	11084	17072	11086
aeb	female	5988	5988	17072	11086
ar	default	12638	12638	23105	12638
ar	female	10467	10467	23105	12638
arq	default	11084	11084	17072	11086
arq	female	5988	5988	17072	11086
ary	default	11084	11084	17072	11086
ary	female	5988	5988	17072	11086
en-simple	default	12602	12602	12602	12602
es	default	12662	12662	12865	12662
es	female	203	203	12865	12662
fa	default	12732	12732	12732	12732
prs	default	12711	12711	12714	12711
prs	female	3	3	12714	12711
ru	default	11084	11084	11090	11084
ru	female	6	6	11090	11084
sq	default	12736	12736	12752	12736
sq	female	16	16	12752	12736
ti	default	12711	12711	24390	12711
ti	female	11679	11679	24390	12711
uk	default	11076	11076	11080	11076
uk	female	4	4	11080	11076

---

Example

French: Avez-vous des nausées ou des vomissements ?
English: Do you have nausea or vomiting?
UMLS gloss: You | Nausea | or – article | Vomiting | Question

---

Citation

If you use the translations, please cite:

@inproceedings{bouillon-etal-2021-speech,
    title = "A Speech-enabled Fixed-phrase Translator for Healthcare Accessibility",
    author = "Bouillon, Pierrette  and
      Gerlach, Johanna  and
      Mutal, Jonathan  and
      Tsourakis, Nikos  and
      Spechbach, Herv{\'e}",
    editor = "Field, Anjalie  and
      Prabhumoye, Shrimai  and
      Sap, Maarten  and
      Jin, Zhijing  and
      Zhao, Jieyu  and
      Brockett, Chris",
    booktitle = "Proceedings of the 1st Workshop on NLP for Positive Impact",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.nlp4posimpact-1.15/",
    doi = "10.18653/v1/2021.nlp4posimpact-1.15",
    pages = "135--142"
}

@article{info:doi/10.2196/13167,
    author="Spechbach, Herv{\'e}
    and Gerlach, Johanna
    and Mazouri Karker, Sanae
    and Tsourakis, Nikos
    and Combescure, Christophe
    and Bouillon, Pierrette",
    title="A Speech-Enabled Fixed-Phrase Translator for Emergency Settings: Crossover Study",
    journal="JMIR Med Inform",
    year="2019",
    month="May",
    day="07",
    volume="7",
    number="2",
    pages="e13167",
    keywords="anamnesis; emergencies; tools for translation and interpreting; fixed-phrase translator; speech modality",
    issn="2291-9694",
    doi="10.2196/13167",
    url="http://medinform.jmir.org/2019/2/e13167/",
    url="https://doi.org/10.2196/13167",
    url="http://www.ncbi.nlm.nih.gov/pubmed/31066702"
}

---

Acknowledgments

This corpus was developed in the context of the BabelDr and PictoDr projects at the University of Geneva, in collaboration with Geneva University Hospitals. This work is part of the PROPICTO project, funded by the Swiss National Science Foundation (N°197864) and the French National Research Agency (ANR-20-CE93-0005). This project also received funding by the ”Fondation Privée des Hôpitaux Universitaires de Genève”.