---
title: "German-Arabic Speech-to-Speech Translation for Psychiatric"
description: "The paper describes the Arabic-side NLP components of RELATER, a turn-based German–Arabic… 95% accuracy. 28 references."
authors: ["Juan Hussain", "Mohammed Mediani", "Moritz Behr", "M. Amin Cheragui", "Sebastian Stüker", "Alexander Waibel"]
year: 2020
venue: "Proceedings of the Fifth Arabic Natural Language Processing Workshop"
source: "https://aclanthology.org/2020.wanlp-1.1/"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/hussain-2020-german-arabic-speech-speech-translation/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/hussain-2020-german-arabic-speech-speech-translation/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/10/) / German-Arabic Speech-to-Speech Translation for Psychiatric Diagnosis

# German-Arabic Speech-to-Speech Translation for Psychiatric Diagnosis

**Authors:** Juan Hussain, [Mohammed Mediani](https://md.research.elamur.ai/free/en/authors/mohammed-mediani/), Moritz Behr, M. Amin Cheragui, [Sebastian Stüker](https://md.research.elamur.ai/free/en/authors/sebastian-stuker/) and [Alexander Waibel](https://md.research.elamur.ai/free/en/authors/alex-waibel/) · **Year:** 2020 · **Venue:** [Proceedings of the Fifth Arabic Natural Language Processing Workshop](https://md.research.elamur.ai/free/en/venues/wanlp/) ([WANLP 2020](https://md.research.elamur.ai/free/en/venues/wanlp/2020/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2020.wanlp-1.1/) · [PDF](https://aclanthology.org/2020.wanlp-1.1.pdf) · [OpenAlex](https://openalex.org/W3120727700)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes the Arabic-side NLP components of RELATER, a turn-based German–Arabic speech-to-speech translation system. It is meant to stand in for scarce human interpreters when German psychiatrists give Arabic-speaking refugees the structured M.I.N.I. diagnostic interview. The authors chose a cascade over an end-to-end model because cascades still perform better and no German–Arabic speech-to-speech corpus exists. The cascade has four parts:
- an LSTM encoder/decoder ASR with attention,
- Transformer-based MT,
- a Flair BiLSTM-CRF tagger that restores Arabic diacritics,
- Tacotron 2 with a WaveGlow vocoder for TTS.

The ASR is trained on Aljazeera broadcast speech and performs poorly on interview speech. The authors recorded a few hours of in-domain audio and adapted the model with it, which more than halved the in-domain WER. The diacritizer, trained on filtered Al-Shamela text, reaches about 95% accuracy. All seven components fit on one GPU and run well faster than real time. The work shows that a small amount of targeted in-domain data can make a niche, low-resource deployment practical.

## Key points
- **ASR setup:** a two-layer CNN front end (32 channels, 4× downsampling), 6 LSTM encoder layers, 2 decoder layers, 40 log-Mel features, 4k BPE units learned on MSA text, and beam size 4.
- **ASR training and baseline:** trained on Alj.1200h. WER is 18.8% on Alj.MSA+dialect.10h and 12.6% on Alj.MSA.2h, but rises to 40.0% on mini-ans.42m and 30.4% on mini.ques.50m.
- **Domain adaptation:** the in-domain set mini.que.ans.3.34h was recorded with TEQST and corrected with DaC-ToR.
  - Mixed fine-tuning lowered in-domain WER to 16.4% and 8.6% and slightly improved the broadcast sets.
  - Decoder-only fine-tuning on domain data reached 6.2% on questions but degraded the broadcast sets to 25.7% and 20.5%.
- **Diacritization:** Al-Shamela (about 1B words) was filtered to lines where more than 40% of characters are short vowels, giving about 5M lines. About 1.5M lines were used for training, and the tagger scored 95.34% accuracy.
- **Latency:** the whole pipeline needs 0.17× the input duration on one TITAN RTX. TTS is the slowest component (4.31 s for 54 words), and ASR, MT and diacritization use 1.4, 0.9 and 0.8 GB of GPU memory.

## Interesting ideas
- The test-domain vocabulary overlapped the training data almost fully (3% OOV), and adapting only the decoder already helped a lot; together these pin the errors on domain mismatch rather than missing words.
- Blending the large general corpus with the small domain set avoided forgetting, unlike pure domain tuning, which collapsed general performance; reporting both in-domain and out-of-domain scores exposes that trade-off.
- The authors flag that their best in-domain number is inflated because training and test shared the same questions, read by different speakers.
- A noisy huge corpus was turned into clean supervision by keeping only lines dense in the target annotation (short vowels above 40% of characters).
- Information that the written form drops (vowels) is restored by a separate explicit tagging step before synthesis, rather than expecting TTS to guess it.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Before blaming a parser's lexicon for errors on a new genre, measure unknown-lemma rates and retrain single components, to separate vocabulary gaps from structural mismatch.
- **I2** (learn): When adapting the perceptron parser or gparse to fables or school tests, train on a mixture and always report a general UD treebank score as a forgetting control.
- **I3** (answer): Keep QA test items disjoint from training texts, not just from training readers or paraphrases, or treat overlapping items as a known-optimistic control.
- **I4** (tokens): Filter silver morphology/lemma data by how completely each sentence is annotated, keeping only densely marked lines for training the tagger.
- **I5** (back): Before realizing text from the meaning graph, add a dedicated step that fills in features the surface form needs (agreement, articles, tense marking), so generation stays inspectable.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 519 works

## Cite

Cite as: Juan Hussain, Mohammed Mediani, Moritz Behr, M. Amin Cheragui, Sebastian Stüker and Alexander Waibel. 2020. German-Arabic Speech-to-Speech Translation for Psychiatric Diagnosis. Proceedings of the Fifth Arabic Natural Language Processing Workshop.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{hussain2020german,
  title = {{German-Arabic Speech-to-Speech Translation for Psychiatric Diagnosis}},
  author = {Juan Hussain, Mohammed Mediani, Moritz Behr, M. Amin Cheragui, Sebastian Stüker and Alexander Waibel},
  booktitle = {Proceedings of the Fifth Arabic Natural Language Processing Workshop},
  year = {2020},
  doi = {10.5445/ir/1000166143},
  url = {https://doi.org/10.5445/ir/1000166143}
}
```

- [← previous From Epic Characters to a Movie Transformation: Tracing the Journey of Beowulf from Old English Poetry to Contemporary Motion Picture](https://md.research.elamur.ai/free/en/papers/gautam-2020-epic-characters-movie-transformation-tracing/)
- [next → Grammaticalization of qәl ‘gourd’ in Amharic](https://md.research.elamur.ai/free/en/papers/gensler-2020-grammaticalization-ql-gourd-amharic/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/10/)

## References

28 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Hamza Abbad and Shengwu Xiong. 2020. Multi-components system for automatic arabic diacritization. In European Conference on Information Retrieval, pages 341–355. Springer. [find](https://scholar.google.com/scholar?q=Hamza+Abbad+and+Shengwu+Xiong%2E+2020%2E+Multi%2Dcomponents+system+for+automatic+arabic+diacritization%2E+In+European+Conference+on+Information+Retrieval%2C+pages+341%E2%80%93355%2E+Springer%2E)
2. Ahmed Abdelali, Francisco Guzman, Hassan Sajjad, and Stephan Vogel. 2014. [The amara corpus: Building parallel language resources for the educational domain](https://md.research.elamur.ai/free/en/papers/abdelali-2014-amara-corpus-building-parallel-language/). In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Thierry Declerck, Hrafn Loftsson, Bente Maegaard, Joseph Mariani, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, editors, Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), Reykjavik, Iceland, may. European Language Resources Association (ELRA).
3. Željko Agić and Ivan Vulić. 2019. [JW300: A wide-coverage parallel corpus for low-resource languages](https://md.research.elamur.ai/free/en/papers/agic-2019-jw300-wide-coverage-parallel-corpus/). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3204–3210, Florence, Italy, July. Association for Computational Linguistics.
4. Alan Akbik, Duncan Blythe, and Roland Vollgraf. 2018. [Contextual string embeddings for sequence labeling](https://md.research.elamur.ai/free/en/papers/akbik-2018-contextual-string-embeddings-sequence-labeling/). In COLING 2018, 27th International Conference on Computational Linguistics, pages 1638–1649.
5. Ahmed Ali, Peter Bell, James Glass, Yacine Messaoui, Hamdy Mubarak, Steve Renals, and Yifan Zhang. 2016. [The mgb-2 challenge: Arabic multi-dialect broadcast media recognition](https://md.research.elamur.ai/free/en/papers/ali-2016-mgb-2-challenge-arabic-multi/). In 2016 IEEE Spoken Language Technology Workshop (SLT), pages 279–284. IEEE.
6. Ebrahim Ansari, Amittai Axelrod, Nguyen Bach, Ondřej Bojar, Roldano Cattoni, Fahim Dalvi, Nadir Durrani, Marcello Federico, Christian Federmann, Jiatao Gu, Fei Huang, Kevin Knight, Xutai Ma, Ajay Nagesh, Matteo Negri, Jan Niehues, Juan Pino, Elizabeth Salesky, Xing Shi, Sebastian Stüker, Marco Turchi, Alexander Waibel, and Changhan Wang. 2020. [find](https://scholar.google.com/scholar?q=Ebrahim+Ansari%2C+Amittai+Axelrod%2C+Nguyen+Bach%2C+Ondr%CC%8Cej+Bojar%2C+Roldano+Cattoni%2C+Fahim+Dalvi%2C+Nadir+Durrani%2C+Marcello+Federico%2C+Christian+Federmann%2C+Jiatao+Gu%2C+Fei+Huang%2C+Kevin+Knight%2C+Xutai+Ma%2C+Ajay+Nag)
7. FINDINGS OF THE IWSLT 2020 EVALUATION CAMPAIGN. In Proceedings of the 17th International Conference on Spoken Language Translation, pages 1–34, Online, July. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=FINDINGS+OF+THE+IWSLT+2020+EVALUATION+CAMPAIGN%2E+In+Proceedings+of+the+17th+International+Conference+on+Spoken+Language+Translation%2C+pages+1%E2%80%9334%2C+Online%2C+July%2E+Association+for+Computational+Linguistics%2E)
8. Yonatan Belinkov, Alexander Magidow, Maxim Romanov, Avi Shmidman, and Moshe Koppel. 2016. [Shamela: A large-scale historical arabic corpus](https://md.research.elamur.ai/free/en/papers/belinkov-2016-shamela-large-scale-historical-arabic/). arXiv preprint arXiv:1612.08989. [↗](https://arxiv.org/abs/1612.08989)
9. Mauro Cettolo, Christian Girardi, and Marcello Federico. 2012. Wit 3 : Web inventory of transcribed and translated talks. In Proceedings of the 16 th Conference of the European Association for Machine Translation (EAMT), pages 261–268, Trento, Italy, May. [find](https://scholar.google.com/scholar?q=Mauro+Cettolo%2C+Christian+Girardi%2C+and+Marcello+Federico%2E+2012%2E+Wit+3+%3A+Web+inventory+of+transcribed+and+translated+talks%2E+In+Proceedings+of+the+16+th+Conference+of+the+European+Association+for+Machine)
10. Chenhui Chu, Raj Dabre, and Sadao Kurohashi. 2017. [An empirical comparison of domain adaptation methods for neural machine translation](https://md.research.elamur.ai/free/en/papers/chu-2017-empirical-comparison-domain-adaptation-methods/). In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 385–391.
11. III Hal Daumé and Daniel Marcu. 2006. Domain Adaptation for Statistical Classifiers. J. Artif. Int. Res., 26(1):101–126, May. [↗](https://doi.org/10.1613/jair.1872)
12. Ali Farghaly and Khaled Shaalan. 2009. Arabic natural language processing: Challenges and solutions. ACM Transactions on Asian Language Information Processing, 8(4), December. [↗](https://doi.org/10.1145/1644879.1644881)
13. Philip Gage. 1994. A new algorithm for data compression. C Users Journal, 12(2):23–38. [find](https://scholar.google.com/scholar?q=Philip+Gage%2E+1994%2E+A+new+algorithm+for+data+compression%2E+C+Users+Journal%2C+12%282%29%3A23%E2%80%9338%2E)
14. Nawar Halabi. 2016. Modern standard arabic phonetics for speech synthesis. Ph.D. thesis, University of Southampton. [find](https://scholar.google.com/scholar?q=Nawar+Halabi%2E+2016%2E+Modern+standard+arabic+phonetics+for+speech+synthesis%2E+Ph%2ED%2E+thesis%2C+University+of+Southampton%2E)
15. Zhiheng Huang, Wei Xu, and Kai Yu. 2015. [Bidirectional lstm-crf models for sequence tagging](https://md.research.elamur.ai/free/en/papers/huang-2015-bidirectional-lstm-crf-models-sequence/). arXiv preprint arXiv:1508.01991. [↗](https://arxiv.org/abs/1508.01991)
16. Juan Hussain, Oussama Zenkri, Sebastian Stüker, and Alex Waibel. 2020. Dactor: A data collection tool for the relater project. In Proceedings of The 12th Language Resources and Evaluation Conference, pages 6627–6632. [find](https://scholar.google.com/scholar?q=Juan+Hussain%2C+Oussama+Zenkri%2C+Sebastian+Stu%CC%88ker%2C+and+Alex+Waibel%2E+2020%2E+Dactor%3A+A+data+collection+tool+for+the+relater+project%2E+In+Proceedings+of+The+12th+Language+Resources+and+Evaluation+Conference%2C)
17. Mohammed Mediani, Joshua Winebarger, and Alexander Waibel. 2014. Improving In-Domain Data Selection for Small In-Domain Sets. [↗](https://doi.org/10.5445/ir/1000166290)
18. Robert C. Moore and William D. Lewis. 2010. [Intelligent Selection of Language Model Training Data](https://md.research.elamur.ai/free/en/papers/moore-2010-intelligent-selection-language-model-training/). In ACL (Short Papers), pages 220–224.
19. Thai-Son Nguyen, Ngoc-Quan Pham, Sebastian Stueker, and Alex Waibel. 2020a. High performance sequence-to-sequence model for streaming speech recognition. arXiv preprint arXiv:2003.10022. [↗](https://arxiv.org/abs/2003.10022)
20. Thai-Son Nguyen, Sebastian Stüker, Jan Niehues, and Alex Waibel. 2020b. Improving sequence-to-sequence speech recognition training with on-the-fly data augmentation. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 7689–7693. IEEE. [find](https://scholar.google.com/scholar?q=Thai%2DSon+Nguyen%2C+Sebastian+Stu%CC%88ker%2C+Jan+Niehues%2C+and+Alex+Waibel%2E+2020b%2E+Improving+sequence%2Dto%2Dsequence+speech+recognition+training+with+on%2Dthe%2Dfly+data+augmentation%2E+In+ICASSP+2020%2D2020+IEEE+Internat)
21. Aaron van den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, and Koray Kavukcuoglu. 2016. [Wavenet: A generative model for raw audio](https://md.research.elamur.ai/free/en/papers/oord-2016-wavenet-generative-model-raw-audio/). arXiv preprint arXiv:1609.03499. [↗](https://arxiv.org/abs/1609.03499)
22. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. [Bleu: a method for automatic evaluation of machine translation](https://md.research.elamur.ai/free/en/papers/papineni-2002-bleu-method-automatic-evaluation-machine/). In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA, July. Association for Computational Linguistics.
23. Ryan Prenger, Rafael Valle, and Bryan Catanzaro. 2019. [Waveglow: A flow-based generative network for speech synthesis](https://md.research.elamur.ai/free/en/papers/prenger-2018-waveglow-flow-based-generative-network/). In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 3617–3621. IEEE.
24. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2015. [Neural machine translation of rare words with subword units](https://md.research.elamur.ai/free/en/papers/sennrich-2016-neural-machine-translation-rare-words/). arXiv preprint arXiv:1508.07909. [↗](https://arxiv.org/abs/1508.07909)
25. David V. Sheehan, Yves Lecrubier, K. Harnett Sheehan, Patricia Amorim, Juris Janavs, Emmanuelle Weiller, Thierry Hergueta, Roxy Baker, and Geoffrey C. Dunbar. 1998. The mini-international neuropsychiatric interview (m.i.n.i.): The development and validation of a structured diagnostic psychiatric interview for dsm-iv and icd-10. The Journal of Clinical Psychiatry, 59(20):22–33. [find](https://scholar.google.com/scholar?q=David+V%2E+Sheehan%2C+Yves+Lecrubier%2C+K%2E+Harnett+Sheehan%2C+Patricia+Amorim%2C+Juris+Janavs%2C+Emmanuelle+Weiller%2C+Thierry+Hergueta%2C+Roxy+Baker%2C+and+Geoffrey+C%2E+Dunbar%2E+1998%2E+The+mini%2Dinternational+neuropsychia)
26. Jonathan Shen, Ruoming Pang, Ron J Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, Rj Skerrv-Ryan, et al. 2018. Natural tts synthesis by conditioning wavenet on mel spectrogram predictions. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4779–4783. IEEE. [find](https://scholar.google.com/scholar?q=Jonathan+Shen%2C+Ruoming+Pang%2C+Ron+J+Weiss%2C+Mike+Schuster%2C+Navdeep+Jaitly%2C+Zongheng+Yang%2C+Zhifeng+Chen%2C+Yu+Zhang%2C+Yuxuan+Wang%2C+Rj+Skerrv%2DRyan%2C+et+al%2E+2018%2E+Natural+tts+synthesis+by+conditioning+wavenet+)
27. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017a. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems 30, pages 5998–6008. Curran Associates, Inc.
28. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017b. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems 30, pages 5998–6008. Curran Associates, Inc.

*Updated 2 November 2026*