Thesis topics for students of Finnish

Looking for a topic for a bachelor’s or master’s thesis? Below are six suggestions. Each one gives a research question, the data it needs, whether that data exists yet and methods that suit it.

Status on 12 October 2026. The game’s own research data is still to come. The measurement system is still at the design stage, and there is no date for its completion. The corpus part of topic 3 and the groundwork for topic 6 can be started now with material from the Language Bank of Finland and Kotus. The other topics wait for data from the game, and topics 4 and 5 also need a separate decision.

The game in brief

In Sanamahti a player finds Finnish words on a letter grid by swiping across adjacent letters. The game accepts words only in their base form. The game is meant to become an instrument that collects three kinds of information: which words players say they know, which strings they offer as words and how an automated editor decides on new words. The method is described on its own page.

Topics

1. Word familiarity and frequency in text

Research question. How well does the frequency of a word in news text predict how many players know it? Which words are better or less well known than their frequency suggests?

Data. Measured familiarity per word from Sanamahti, and lemma frequencies from a text corpus, for example from the Language Bank of Finland (Kielipankki).

Does the data exist? The familiarity data is planned. Text corpora are available in the Language Bank of Finland.

Methods. Correlation and regression analysis, and a qualitative look at the outliers by word class and register.

2. Inflected forms offered as base forms

Research question. Which inflected forms do players swipe even though the game accepts only base forms? Do the most common ones point to lexicalisation, in which an inflected form has started to work as an independent word? The adverb kotona (“at home”) and the plural-only noun sakset (“scissors”), for example, are already headwords in the Kotus word list.

Data. Rejected swipes for which the morphological analyser has given a lemma and a description of the form, with the number of swipes.

Does the data exist? Planned.

Methods. Classifying the forms by case, number and word class, comparing them with dictionary headwords and corpus frequencies, and a qualitative analysis of case forms that have become established as adverbs and of plural-only nouns.

3. Two spellings of a loanword

Research question. How are the foreign spelling and the adapted Finnish spelling of a loanword, for example pizza and pitsa, distributed in news text over the years? Which spelling do players know better?

Data. Yearly frequencies of the parallel spellings from a text corpus, information on which spellings are dictionary headwords, and from Sanamahti the familiarity of both spellings.

Does the data exist? The corpus part can be started now. Yle’s news archive is in the Language Bank of Finland. Check the terms of use of the material there. The Kotus word list is open, and both pizza and pitsa are headwords in it. The Sanamahti familiarity data is planned. In the design, the familiarity strip gets the words whose status the automated editor is least certain about, so measuring an established pair of spellings has to be agreed separately. Swipes cannot be compared: the design counts only rejected swipes, and the game accepts both pizza and pitsa. The board also limits what can be swiped.

Methods. Diachronic corpus analysis, a time series of the relative shares of the spellings, and a description of how loanwords are adapted, from unadapted loans to fully adapted ones.

4. Reliability of the pseudo-word control in a game

Research question. How reliably does a yes/no vocabulary test measure word knowledge when it is taken inside a game and for a reward? Do pseudo-words separate careful respondents from guessers?

Data. Response-level familiarity ratings: the respondent’s pseudonym, the word, the response and the responses to pseudo-words.

Does the data exist? Planned, and it also needs a decision. The design records response counts per word, per-player calibration data from pseudo-word hits and a pseudonymous record of which word a player has rated. It does not yet record the value of an individual response (know or don’t know) or the responses to individual pseudo-words, so say early if you need them. Response-level data is not intended to be published openly. Using it in research would require a separate agreement and a data protection assessment, and no decision on that has been made.

Methods. Signal detection theory (hits and false alarms), a comparison of alternative guessing corrections, reliability estimation and item response theory.

5. The spread of student slang between universities

Research question. Is a student slang word known equally well in different universities and student associations? Does the data show a word spreading from one community to another?

Data. Word familiarity and word proposals by university community.

Does the data exist? No, and the current design does not include community-level data. The game’s daily challenge has leaderboards for universities and student associations, but the design does not link word observations to communities. Linking them would require a separate decision and a data protection assessment, because the communities can be small.

Methods. Variationist sociolinguistics, comparison between communities and time series. The data can be complemented with a survey or interviews.

6. Does automated editing anticipate the dictionary?

Research question. How often is a word accepted by Sanamahti’s automated editor later added to Kielitoimiston sanakirja, the Kotus dictionary of contemporary Finnish? On what kind of evidence do the automated editor and the dictionary editors reach different outcomes?

Data. Sanamahti’s decision log with its evidence, and later updates of the dictionary. The yearly selections of topical words published by Kotus (sanapoiminnot) and a news corpus are useful points of comparison.

Does the data exist? The decision log is planned. The dictionary is updated every year or two, so a comparison with future additions is a long design and suits a master’s thesis better than a bachelor’s thesis. As groundwork, it is already possible to study how the new words in the Kotus selections appear in a news corpus over the years.

Methods. Retrospective corpus analysis, research on how new words become established, and classification and error analysis of the decisions.

What a student gets

• A documented method. The planned way of collecting the data, its limitations and the open questions are written out on the method page. You can cite the description and evaluate it.
• A direct line to the author. Sanamahti is made by one developer, Alex Liao. He answers questions about how the data comes about.
• The data, when it exists. When the open data is published, it is available to everyone. If your topic needs something else, it has to be agreed separately.
• A chance to influence. The system is still at the design stage. If your topic needs a measurement that the design does not include, say so early.

What a supervisor gets

• A description of how the data is collected and of its biases, before any data exists.
• Topics that can be narrowed to a bachelor’s thesis or extended to a master’s thesis.
• In time, open data with no player identifiers.
• A contact person who knows how the system is implemented.

The topics can be adapted and offered to students freely.

What Sanamahti does not offer

Supervision comes from your own university. Sanamahti is the project of one developer and cannot give linguistic supervision. The work is yours: you define the topic with your supervisor, and Sanamahti has no influence on the results. This is not a commission or paid work.

There is no date for when the game’s data will be ready. Do not tie your graduation to it unless your topic also works with other data.

Get in touch

Write through the contact form and choose student as your role. Say which topic interests you, what you study and when you plan to start. A supervisor can use the same form. The other pages for researchers are linked from For researchers.