---
title: "Multi-view Dimensionality Reduction for Dialect Identification of"
description: "The paper addresses identifying the dialect of spoken Arabic (DID), distinguishing five classes: Modern… 19 references, cited by 3 papers here."
authors: ["Sameer Khurana", "Ahmed Ali", "Steve Renals"]
year: 2016
venue: "arXiv:1609.05650"
source: "https://arxiv.org/abs/1609.05650"
doi: "10.48550/arxiv.1609.05650"
arxiv: "1609.05650"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/khurana-2016-multi-view-dimensionality-reduction-dialect/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/khurana-2016-multi-view-dimensionality-reduction-dialect/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2016](https://md.research.elamur.ai/free/en/papers/2016/page/5/) / Multi-view Dimensionality Reduction for Dialect Identification of Arabic Broadcast Speech

# Multi-view Dimensionality Reduction for Dialect Identification of Arabic Broadcast Speech

**Authors:** [Sameer Khurana](https://md.research.elamur.ai/free/en/authors/sameer-khurana/), [Ahmed Ali](https://md.research.elamur.ai/free/en/authors/ahmed-m-abdelatty-ali/) and [Steve Renals](https://md.research.elamur.ai/free/en/authors/steve-renals/) · **Year:** 2016 · **Venue:** arXiv:1609.05650 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/1609.05650) · [PDF](https://arxiv.org/pdf/1609.05650) · [DOI 10.48550/arxiv.1609.05650](https://doi.org/10.48550/arxiv.1609.05650) · [arXiv 1609.05650](https://arxiv.org/abs/1609.05650) · [OpenAlex](https://openalex.org/W2523233994)

**Citations:** 2 citations (OpenAlex, 2026-10-14) · [3](https://md.research.elamur.ai/free/en/papers/khurana-2016-multi-view-dimensionality-reduction-dialect/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper addresses identifying the dialect of spoken Arabic (DID), distinguishing five classes: Modern Standard Arabic (MSA) plus the Egyptian (EGY), Gulf (GLF), North African (NOR) and Levantine (LAV) varieties. Such systems usually rely on two separate utterance representations. A phonotactic one counts phone n-grams from a phone recogniser. An acoustic one uses i-Vectors built on bottleneck features (BNF). Normally each representation trains its own classifier and the scores are fused afterwards. The authors instead merge the two views into a single vector space. They use Canonical Correlation Analysis (CCA), which finds projections of each view that correlate maximally with each other, and concatenate the two projections. The data is broadcast speech, with roughly 9–13 training hours per dialect and test material from Al-Jazeera. The fused CCA space beats either view alone. After LDA+WCCN it matches the processed acoustic space, and combining the two gives the best accuracy. The approach matters because it needs only one back-end classifier. CCA is also unsupervised, so in principle it can absorb unlabeled out-of-domain data.

## Key points
- The phonotactic VSM is a term-document matrix over phone 2- and 3-grams, about 8K features, reduced by truncated SVD to 1200 dimensions.
- The acoustic VSM consists of 400-dimensional i-Vectors from a GMM-UBM with 2048 Gaussians, trained on DNN bottleneck features (BNF).
- CCA projects each view to 300 dimensions, and concatenating them gives a 600-dimensional shared space. LDA and WCCN are then applied.
- Accuracy results:

| Representation | Accuracy |
|---|---|
| Phonotactic | 0.45 |
| Acoustic | 0.55 |
| Raw CCA | 0.56 |
| CCA+LDA+WCCN | 0.58 |
| Acoustic+LDA+WCCN | 0.58 |
| The two combined (8 dims) | 0.60 (best) |

- Training uses 8,440 utterances and testing 1,521. MSA is the most clearly separated dialect, while EGY, GLF, LAV and NOR are frequently mistaken for one another, LAV especially.

## Interesting ideas
- Two heterogeneous representations of the same item (symbol-sequence counts vs. continuous embeddings) can be fused by projecting each onto directions where they agree most, then concatenating the projections, rather than voting at the score level.
- Feature-level fusion lost some discriminative signal, and adding back the single strongest view gave the best result, so fusion should be checked against the best individual view rather than assumed superior.
- Treating sequences of discrete symbols as documents of n-gram "terms" and compressing them with truncated SVD gives a cheap, linear, interpretable profile of a style or variety.
- Because CCA needs no labels, the authors note it could take in unlabeled data from another domain, acting like co-training or domain adaptation (proposed, not yet tested).
- Pairing a confusion matrix with a low-dimensional projection of the learned space lets one explain which classes collide and why.

## For Mova

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

- **I1** (learn): when several witness parsers produce different feature views of a sentence, CCA could build a shared space where their agreement is explicit and inspectable, helping select silver data.
- **I2** (learn): any merged witness representation should be evaluated alongside its strongest single component as a control.
- **I3** (tree): n-gram profiles over UPOS/deprel sequences could fingerprint which treebank convention or UD dialect an input follows before choosing a conversion path.
- **I4** (learn): unlabeled English text seen through two views, such as tagger output and parser output, could adapt shared representations to new genres without annotation.
- **I5** (answer): projecting question or frame types and inspecting confusions could explain where readers systematically mix up categories, keeping errors explainable.

*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 (7): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,552 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 870 works · [🔊 Signal Processing](https://md.research.elamur.ai/free/en/shelves/computer-science/signal-processing/) · 452 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Sameer Khurana, Ahmed Ali and Steve Renals. 2016. Multi-view Dimensionality Reduction for Dialect Identification of Arabic Broadcast Speech. arXiv:1609.05650.

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

```bibtex
@misc{khurana2016multi,
  title = {{Multi-view Dimensionality Reduction for Dialect Identification of Arabic Broadcast Speech}},
  author = {Sameer Khurana, Ahmed Ali and Steve Renals},
  year = {2016},
  doi = {10.48550/arxiv.1609.05650},
  eprint = {1609.05650},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1609.05650}
}
```

- [← previous Modelling the Usage of Discourse Connectives as Rational Speech Acts](https://md.research.elamur.ai/free/en/papers/yung-2016-modelling-usage-discourse-connectives-as/)
- [next → Predicting and Using a Pragmatic Component of Lexical Aspect of Simple Past Verbal Tenses for Improving English-to-French Machine Translation](https://md.research.elamur.ai/free/en/papers/loaiciga-2016-predicting-pragmatic-component-lexical-aspect/)
- [↑ back to papers of 2016](https://md.research.elamur.ai/free/en/papers/2016/page/5/)

## References

19 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Fadi Biadsy, Automatic dialect and accent recognition and its application to speech recognition, Ph.D. thesis, Columbia University, 2011. [find](https://scholar.google.com/scholar?q=Fadi+Biadsy%2C+Automatic+dialect+and+accent+recognition+and+its+application+to+speech+recognition%2C+Ph%2ED%2E+thesis%2C+Columbia+University%2C+2011%2E)
2. Ahmed Ali, Najim Dehak, Patrick Cardinal, Sameer Khurana, Sree Harsha Yella, James Glass, Peter Bell, and Steve Renals, “[Automatic dialect detection in arabic broadcast speech](https://md.research.elamur.ai/free/en/papers/ali-2015-automatic-dialect-detection-arabic-broadcast/),” in Interspeech 2016, 2016, pp. 2934–2938.
3. Haizhou Li, Bin Ma, and Chin-Hui Lee, “A vector space modeling approach to spoken language identification,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 15, no. 1, pp. 271–284, 2007. [↗](https://doi.org/10.1109/tasl.2006.876860)
4. Scott Deerwester, Susan T Dumais, George W Furnas, Thomas K Landauer, and Richard Harshman, “Indexing by latent semantic analysis,” Journal of the American society for information science, vol. 41, no. 6, pp. 391, 1990. [find](https://scholar.google.com/scholar?q=Scott+Deerwester%2C+Susan+T+Dumais%2C+George+W+Furnas%2C+Thomas+K+Landauer%2C+and+Richard+Harshman%2C+%E2%80%9CIndexing+by+latent+semantic+analysis%2C%E2%80%9D+Journal+of+the+American+society+for+information+science%2C+vol%2E+41%2C+no)
5. Bruce Moore, “Principal component analysis in linear systems: Controllability, observability, and model reduction,” IEEE transactions on automatic control, vol. 26, no. 1, pp. 17–32, 1981. [↗](https://doi.org/10.1109/tac.1981.1102568)
6. Marc A Zissman and Kay M Berkling, “Automatic language identification,” Speech Communication, vol. 35, no. 1, pp. 115–124, 2001. [↗](https://doi.org/10.1016/s0167-6393(00)00099-6)
7. Fadi Biadsy, Julia Hirschberg, and Nizar Habash, “[Spoken arabic dialect identification using phonotactic modeling](https://md.research.elamur.ai/free/en/papers/biadsy-2009-spoken-arabic-dialect-identification-phonotactic/),” in Proceedings of the eacl 2009 workshop on computational approaches to semitic languages. Association for Computational Linguistics, 2009, pp. 53–61.
8. Fadi Biadsy, Julia Hirschberg, and Daniel PW Ellis, “Dialect and accent recognition using phonetic-segmentation supervectors.,” in INTERSPEECH, 2011, pp. 745–748. [↗](https://doi.org/10.7916/d8p84mcw)
9. David Martı́nez, Lukáš Burget, Luciana Ferrer, and Nicolas Scheffer, “ivector-based prosodic system for language identification,” in 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2012, pp. 4861–4864. [find](https://scholar.google.com/scholar?q=David+Mart%C4%B1%CC%81nez%2C+Luka%CC%81s%CC%8C+Burget%2C+Luciana+Ferrer%2C+and+Nicolas+Scheffer%2C+%E2%80%9Civector%2Dbased+prosodic+system+for+language+identification%2C%E2%80%9D+in+2012+IEEE+International+Conference+on+Acoustics%2C+Speech+and+Signa)
10. Najim Dehak, Pedro A Torres-Carrasquillo, Douglas A Reynolds, and Reda Dehak, “Language recognition via i-vectors and dimensionality reduction.,” in INTER-SPEECH. Citeseer, 2011, pp. 857–860. [find](https://scholar.google.com/scholar?q=Najim+Dehak%2C+Pedro+A+Torres%2DCarrasquillo%2C+Douglas+A+Reynolds%2C+and+Reda+Dehak%2C+%E2%80%9CLanguage+recognition+via+i%2Dvectors+and+dimensionality+reduction%2E%2C%E2%80%9D+in+INTER%2DSPEECH%2E+Citeseer%2C+2011%2C+pp%2E+857%E2%80%93860%2E)
11. H. Hotelling, “Canonical correlation analysis (cca),” in Journal of Educational Psychology, 1935. [find](https://scholar.google.com/scholar?q=H%2E+Hotelling%2C+%E2%80%9CCanonical+correlation+analysis+%28cca%29%2C%E2%80%9D+in+Journal+of+Educational+Psychology%2C+1935%2E)
12. Patrick Cardinal, Najim Dehak, Yu Zhang, and James Glass, “Speaker adaptation using the i-vector technique for bottleneck features,” Proceedings of Interspeech, vol. 2015, 2015. [find](https://scholar.google.com/scholar?q=Patrick+Cardinal%2C+Najim+Dehak%2C+Yu+Zhang%2C+and+James+Glass%2C+%E2%80%9CSpeaker+adaptation+using+the+i%2Dvector+technique+for+bottleneck+features%2C%E2%80%9D+Proceedings+of+Interspeech%2C+vol%2E+2015%2C+2015%2E)
13. Ahmed Ali, Yifan Zhang, Patrick Cardinal, Najim Dahak, Stephan Vogel, and James Glass, “A complete kaldi recipe for building arabic speech recognition systems,” in Spoken Language Technology Workshop (SLT), 2014 IEEE. IEEE, 2014, pp. 525–529. [find](https://scholar.google.com/scholar?q=Ahmed+Ali%2C+Yifan+Zhang%2C+Patrick+Cardinal%2C+Najim+Dahak%2C+Stephan+Vogel%2C+and+James+Glass%2C+%E2%80%9CA+complete+kaldi+recipe+for+building+arabic+speech+recognition+systems%2C%E2%80%9D+in+Spoken+Language+Technology+Workshop+)
14. Patrick Kenny, Pierre Ouellet, Najim Dehak, Vishwa Gupta, and Pierre Dumouchel, “A study of interspeaker variability in speaker verification,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 16, no. 5, pp. 980–988, 2008. [↗](https://doi.org/10.1109/tasl.2008.925147)
15. Paramveer S Dhillon, Dean P Foster, and Lyle H Ungar, “Eigenwords: Spectral word embeddings,” The Journal of Machine Learning Research, vol. 16, no. 1, pp. 3035– 3078, 2015. [find](https://scholar.google.com/scholar?q=Paramveer+S+Dhillon%2C+Dean+P+Foster%2C+and+Lyle+H+Ungar%2C+%E2%80%9CEigenwords%3A+Spectral+word+embeddings%2C%E2%80%9D+The+Journal+of+Machine+Learning+Research%2C+vol%2E+16%2C+no%2E+1%2C+pp%2E+3035%E2%80%93+3078%2C+2015%2E)
16. David R Hardoon, Sandor Szedmak, and John Shawe-Taylor, “Canonical correlation analysis; an overview with application to learning methods,” 2003. [↗](https://doi.org/10.1162/0899766042321814)
17. Samantha Wray and Ahmed Ali, “Crowdsource a little to label a lot: Labeling a speech corpus of dialectal arabic,” in INTERSPEECH, 2015. [find](https://scholar.google.com/scholar?q=Samantha+Wray+and+Ahmed+Ali%2C+%E2%80%9CCrowdsource+a+little+to+label+a+lot%3A+Labeling+a+speech+corpus+of+dialectal+arabic%2C%E2%80%9D+in+INTERSPEECH%2C+2015%2E)
18. Kevin P Murphy, Machine learning: a probabilistic perspective, MIT press, 2012. [find](https://scholar.google.com/scholar?q=Kevin+P+Murphy%2C+Machine+learning%3A+a+probabilistic+perspective%2C+MIT+press%2C+2012%2E)
19. Bianca Zadrozny and Charles Elkan, “Transforming classifier scores into accurate multiclass probability estimates,” in Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2002, pp. 694–699. [find](https://scholar.google.com/scholar?q=Bianca+Zadrozny+and+Charles+Elkan%2C+%E2%80%9CTransforming+classifier+scores+into+accurate+multiclass+probability+estimates%2C%E2%80%9D+in+Proceedings+of+the+eighth+ACM+SIGKDD+international+conference+on+Knowledge+discov)

*Updated 1 November 2026*