Creative Projects
08 Creative Projects — Building a Computational Linguistics Portfolio
Fatima, the committee highlighted the potential of your Somali‑Bantu dictionary work as something far more powerful than a simple glossary. With careful structure and documentation, it can become a genuine computational linguistics dataset — the kind of project that demonstrates both linguistic curiosity and technical capability. For universities such as MIT and research‑focused programs, a well‑built open dataset can function almost like a mini research lab: it shows that you understand data structure, reproducibility, and how language technology actually gets built.
Your goal over the next 6–9 months should be to transform the dictionary project into a small but credible NLP resource that other researchers could realistically use. The most compelling student portfolios do not just show code — they show tools, datasets, and experiments that others can build on. The following plan outlines how to convert your existing project into a polished research artifact.
Project 1: Somali‑Bantu Open Linguistic Dataset
The strongest version of your project is a structured open dataset rather than a static dictionary. Linguistics and NLP researchers rely heavily on well‑annotated datasets, especially for languages that have limited digital resources. By organizing your work into a machine‑readable dataset with metadata and documentation, you demonstrate both linguistic understanding and technical discipline.
Core Dataset Structure
- Lexicon Table – Somali‑Bantu word, English translation, part‑of‑speech tags, and example sentences.
- Tokenized Text – Short sentences or phrases broken into tokens for NLP training.
- Audio Recordings – Native pronunciation clips paired with written text.
- Alignment Data – Mapping between Somali‑Bantu and English words or phrases.
- Linguistic Metadata – Grammatical notes, morphological information, or dialect notes if relevant.
Each entry should be stored in a structured format such as JSON, CSV, or a simple database. The key idea is reproducibility: someone should be able to download the dataset and immediately use it for computational experiments.
Suggested Technical Stack
- Python for data processing
- Pandas for dataset cleaning and formatting
- JSON or CSV dataset storage
- GitHub for version control and open release
- Hugging Face Datasets for public distribution
You have not provided information about your current programming background, so if Python or data processing tools are new to you, consider starting with basic tutorials before building the full dataset pipeline.
Project 2: Somali‑Bantu Translation Benchmark
Datasets become far more valuable when they include a task researchers can test against. Consider creating a small benchmark for Somali‑Bantu translation or language modeling. Even a modest evaluation task can demonstrate that your dataset has research value.
Possible Benchmark Tasks
- Somali‑Bantu → English translation evaluation set
- Pronunciation recognition dataset using audio recordings
- Sentence alignment challenge for bilingual text
For example, you could release:
- A training dataset
- A validation dataset
- A small held‑out test dataset
You would then include example code showing how someone might train a simple translation or speech model using the dataset. The purpose is not to produce a cutting‑edge model but to demonstrate the workflow used in computational linguistics research.
Project 3: Reproducible NLP Pipeline
Admissions reviewers often look for evidence that a student understands how real research workflows function. A reproducible pipeline signals maturity and technical discipline.
Your repository should include scripts that automatically:
- Clean and standardize the raw dictionary entries
- Tokenize text data for NLP tasks
- Process and label audio files
- Export the final dataset into release format
A simple directory structure might look like this:
- /data_raw – original dictionary entries and recordings
- /scripts – Python processing scripts
- /dataset_release – final cleaned dataset
- /examples – model training examples
- /docs – documentation and tutorials
This structure mirrors the way many open NLP datasets are published. The goal is to make your work transparent and reusable.
Project 4: Research‑Style Documentation
Documentation is often the difference between a hobby project and a research contribution. Your dataset should include clear instructions so that someone unfamiliar with the project can use it easily.
Recommended Documentation Sections
- Overview of the Somali‑Bantu language context
- Description of the dataset structure
- Instructions for loading the dataset
- Tutorial showing how to train a simple translation model
- Explanation of the benchmark task
A strong README file on GitHub is essential. You may also want to include short tutorial notebooks that demonstrate how the dataset can be used in practice.
Publishing Strategy
Once the dataset and documentation are ready, releasing the project publicly is critical. A private or incomplete project does not demonstrate impact in the same way.
Recommended Platforms
- GitHub – full project repository and documentation
- Hugging Face Datasets – structured dataset distribution
Your GitHub repository should include:
- Complete dataset files
- Processing scripts
- Example code for training a model
- Clear documentation
This turns the project into something admissions readers can actually explore, which makes it far more memorable than a traditional activity description.
Portfolio Presentation
By the time applications open, you should have a single portfolio link that demonstrates the full project lifecycle. Consider organizing it into three components:
- Dataset repository
- Benchmark experiment notebook
- Short technical write‑up explaining the design
Even if the dataset is modest in size, the intellectual framing — documenting a low‑resource language dataset with reproducible tools — is exactly the type of initiative that aligns with computational linguistics programs.
Development Timeline (Junior Year → Summer)
| Month | Key Actions |
|---|---|
| February |
|
| March |
|
| April |
|
| May |
|
| June |
|
| July–August |
|
If executed well, this project becomes more than an activity — it becomes a concrete contribution to computational linguistics resources. That combination of linguistic insight, technical infrastructure, and public release is exactly what distinguishes standout portfolios in this field.