PhD Research Proposal Template with Worked Example

When to use this

Many PhD applications, especially in the UK, Australia, China, Japan, Korea, Germany (for individual doctorates) and HEC-funded overseas schemes, ask for a research proposal. It is typically 1,500 to 3,000 words, but always check the limit. The proposal does not lock you into a final project; it shows that you can identify a meaningful question, know the literature, choose suitable methods and plan realistically.

What reviewers look for:

• A clear, focused research question that is worth answering
• Awareness of existing literature and the gap your work addresses
• Methods that are appropriate and feasible within three to four years
• Fit with the supervisor's expertise and the university's resources
• Good academic writing with proper references

The template (with a worked example)

Below, each section has a short explanation followed by an example from a fictional applicant, Fatima Noor, who completed an MS in Computer Science at FAST-NUCES Islamabad and wants to do a PhD on Urdu speech recognition.

1. Title

A specific, informative title, usually 10 to 20 words.

Example: Low-Resource Speech Recognition for Code-Switched Urdu-English Conversation Using Self-Supervised Learning

2. Abstract (150 to 250 words)

Summarise the problem, gap, aims, methods and expected contribution.

Example: Urdu is spoken by more than [number] million people, yet automatic speech recognition (ASR) systems perform poorly on everyday Pakistani speech, which frequently switches between Urdu and English within a single sentence. Existing systems are trained largely on read speech and treat the two languages separately. This project will develop and evaluate ASR models for code-switched Urdu-English conversation using self-supervised pre-training and limited labelled data. It will (1) build a small but carefully annotated corpus of conversational code-switched speech, (2) compare strategies for adapting multilingual self-supervised models to this setting and (3) evaluate performance on downstream tasks such as voice-based information services. The research aims to contribute methods transferable to other low-resource, code-switched languages and a public benchmark for Urdu-English ASR.

3. Introduction and background

Set the scene: why does the topic matter (academically and practically)? Keep it short and lead toward the problem.

Example: Voice interfaces could help millions of Pakistanis with limited literacy access banking, health and government information. However, conversational speech in cities such as Lahore and Karachi routinely mixes Urdu and English ("Main kal meeting attend nahi kar sakti"). Current ASR systems trained on monolingual data handle such speech poorly, leading to high error rates and limited real-world adoption.

4. Literature review and research gap

Summarise the key work in the field and identify precisely what is missing. Organise by theme, not by paper.

Example: Research on low-resource ASR has advanced quickly with self-supervised models such as wav2vec 2.0 and its multilingual variants, which learn speech representations from unlabelled audio [references]. Work on code-switched ASR has focused mainly on Mandarin-English, Hindi-English and Spanish-English [references]. For Urdu, published work has largely used read speech corpora [references], and evaluation on spontaneous code-switched speech remains limited. The gap: there is no systematic study of how self-supervised models can be adapted to conversational Urdu-English speech, and no openly available benchmark for this task.

5. Research questions and objectives

Two to four specific questions. Each should be answerable with your methods.

Example:

• RQ1: How well do existing multilingual self-supervised speech models perform on conversational code-switched Urdu-English speech?
• RQ2: Which adaptation strategies (continued pre-training on Urdu audio, language-aware tokenisation, data augmentation) most improve performance with under 50 hours of labelled data?
• RQ3: How do improvements in word error rate translate into performance on a practical task, such as answering spoken questions about public services?

6. Methodology

Explain the design, data, methods, tools and how you will evaluate results. Address ethics where relevant.

Example:

• Data: Collect around [100] hours of unlabelled conversational audio from consenting volunteers in [three] cities, and annotate [30 to 50] hours, using transcription guidelines developed in the first six months. Supplement with existing public Urdu corpora where licences allow.
• Models: Fine-tune multilingual self-supervised models as baselines; test continued pre-training and tokenisation strategies.
• Evaluation: Word and character error rates on a held-out test set balanced by gender, region and degree of code-switching; task-based evaluation with a prototype question-answering system.
• Ethics: Obtain informed consent, anonymise recordings and seek approval from the university's ethics committee.
• Resources: GPU computing through the university cluster; annotation support budgeted from research funds.

7. Expected contribution and significance

What will be new, and who will benefit?

Example: The project will deliver (1) a benchmark dataset for code-switched Urdu-English speech, (2) evidence on adaptation strategies for low-resource code-switched ASR and (3) a prototype demonstrating practical value. Findings should transfer to other South Asian languages with similar patterns, such as Punjabi-English and Sindhi-Urdu.

8. Timeline

A realistic plan, often as a table.

Example:

Period	Main activities
Year 1	Literature review, annotation guidelines, data collection, baseline experiments (RQ1)
Year 2	Adaptation experiments (RQ2), first conference paper
Year 3	Task-based evaluation (RQ3), journal paper, dataset release
Year 3.5 to 4	Thesis writing and submission

9. Fit with the supervisor and institution

One short paragraph explaining why this group is the right place.

Example: Dr [Name]'s work on [self-supervised learning for low-resource languages] and the group's computing resources make [University] an ideal setting for this research. My MS thesis on [Urdu text classification] and my experience building an Urdu speech dataset of [15] hours provide a solid foundation.

10. References

Use a consistent citation style (APA, IEEE or Harvard). Cite recent, credible sources, and make sure every reference is real.

Example: [Author, A., Author, B. (Year). Title. Journal, volume(issue), pages.] (List 15 to 30 references for a full proposal.)

How to adapt this

• Read the university's guidance first. Some want a two-page outline; others want a full 3,000-word document with set headings.
• Write the research questions before anything else. Every other section should support them.
• Contact potential supervisors early and share a short version. Many will suggest changes that improve your fit.
• Show feasibility. Mention data access, equipment, fieldwork permissions or collaborations, especially if your research involves Pakistan.
• Use a table for the timeline so reviewers can scan it quickly.
• Keep the literature review analytical. Group papers by idea and explain what they show and what they miss.

Common mistakes

• A topic that is far too broad ("Improving education in Pakistan").
• No clear research question, only a description of a topic.
• Methods that are vague ("I will use machine learning") or unrealistic for the time and budget.
• Citing outdated or non-academic sources, or references that do not exist.
• Ignoring the supervisor's area of expertise.
• Copying sections from published papers or other proposals. Plagiarism checks are standard.