Skip to main content
User Image

Dr. Muhammad Swaileh A. Alzaidi

Associate Professor

Advisor to the Vice President for Graduate Studies and Scientific Research – Director of the Research Center, College of Language Sciences – Faculty Member, Department of English

College of Language Sciences
AB2

Modeling Focus Prosody in Hijazi Arabic: A Two-Stage Machine Learning Approach to Classification and Localization

This study investigates how prosodic focus is encoded in Hijazi Arabic using a two-stage machine-learning framework trained on an annotated corpus of scripted speech. We model the predictive contribution of acoustic-prosodic cues: F₀ excursion, intensity, duration, and pitch-level measures (mean F0/max F0/min F0), and, for region-level analyses, phonological metadata (e.g., syllable weight). Stage 1 asks whether an utterance contains focus (Focus vs. Neutral). Evaluated on a natural, imbalanced test set (≈90% Neutral: 10% Focus), a Random Forest achieved Macro-F1 = 0.580 (95% CI: 0.561–0.600) and Accuracy = 0.696 (95% CI: 0.680–0.713), with ROC–AUC = 0.879 and PR–AUC = 0.511. A threshold sweep identified τ ≈ 0.80 as a recall-oriented operating point (P ≈ 0.44, R ≈ 0.61, F1 ≈ 0.51). Feature-importance analysis shows that excursion, intensity, and duration dominate over static pitch height. Stage 2 extends the task to localizing focus regions (On-, Post-, Pre-Focus). Overall performance reaches Macro-F1 ≈ 0.65 (Accuracy ≈ 74.6%), with the Post-Focus region best detected (F1 ≈ 0.85, Recall ≈ 0.87), reflecting robust post-focus compression, while On- and Pre-Focus are weaker and more confusable. Taken together, the results place Hijazi Arabic within a range-driven, +PFC typology: focus is realized by larger F₀ excursions, increased intensity, and lengthening on the focused item, followed by systematic post-focus compression. The approach provides a replicable ML methodology for modeling focus realization in under-resourced dialects and yields findings compatible with PENTA target-dynamics and interpretable within Autosegmental-Metrical accounts.

Publisher Name
Saudi Journal of Language Studies
Volume Number
16-44
more of publication
publications

This study investigates how prosodic focus is encoded in Hijazi Arabic using a two-stage machine-learning framework trained on an annotated corpus of scripted speech.

2026
Published in:
Saudi Journal of Language Studies
publications

Experimental studies show that the prosodic realization of focus varies across Arabic dialects. This study examines focus in Bedouin Hijazi Arabic. Using the question-and-answer paradigm, eleven…

2026
Published in:
Jordan Journal of Modern Languages & Literatures
publications

This study investigates focus intonation in Qassimi Arabic (QA) with two aims: to examine focus-related prosodic changes in the focused word and surrounding regions, and to identify the functional…

2026