Abstract: This work introduces TCBLex, a lexical database of Finnish literary works read by children between the ages of 7 and 15. We explain in detail the work done to build the corpus TCBLex is based on, including how books were sampled and collected, turned into text files, and finally processed. We also touch on legal considerations and how it is possible to build such a corpus in the EU. TCBLex contains over 11 million tokens that are annotated with parts-of-speech tags and lemmatized. We provide 14 different sub-lexicons in total, covering individual intended reading ages, age groups, as well as different genres. We also provide versions with additional morphological information, such as the cases and tenses of words. TCBLex provides various psycholinguistically interesting lexical statistics for both word types and lemmas, such as different frequency metrics, distributions, word lengths, numbers of syllables, morphological paradigm sizes, and for the first time in a Finnish lexicon, ages when words and lemmas are first encountered in books.
Introduction
The development of specialized linguistic resources for children is a critical endeavor for understanding language acquisition and improving educational materials. However, such resources are often scarce for languages other than English. For Finnish, an agglutinative language characterized by high morphological complexity, the need is particularly acute. Standard frequency lists often fail to capture the nuances of Finnish grammar, where a single root can manifest in hundreds of inflected forms through various prefixes and suffixes.
TCBLex addresses this gap as the first comprehensive lexical database of Finnish literary texts curated specifically for children. By providing high-resolution data on word frequencies and morphological scaling, it offers a new lens through which to view the linguistic development of Finnish youth as they progress from simple picture books to complex young adult novels.
Methodology
The construction of TCBLex followed a rigorous multi-stage pipeline to ensure both representative sampling and technical accuracy.
Sampling and Legal Framework
The researchers sampled 300 books—including fiction, non-fiction, and textbooks—based on sales rankings and library loan statistics. To navigate copyright restrictions, the project utilized the European Union’s Text and Data Mining (TDM) exception, which allows research organizations to create digital copies of copyrighted works they have legal access to for scientific purposes.
Digitization and NLP Pipeline
For titles without existing digital versions, the team employed high-resolution photography and Google Cloud Vision’s Document AI for text extraction. The raw text was then processed using Trankit, a state-of-the-art syntactic parser. This allowed for the "lemmatization" of the corpus—grouping complex inflected forms (like tulen, tulemme) under their base dictionary entries (tulla)—with an estimated accuracy of 95–96%.
The TCBLex Dataset
The database is derived from the Turku Children’s Book Corpus (TCBC), which contains approximately 11.6 million tokens. TCBLex itself consists of 14 sub-lexicons, allowing researchers to analyze language usage across specific age groups (ranging from 5 to 15+) and genres.
Key Insights and Results
Analysis of TCBLex has yielded several significant findings regarding the nature of Finnish children's literature:
- Morphological Complexity: A striking 57.5% of word types in the database are hapax legomena (words appearing only once). This high percentage underscores the extreme morphological variety children encounter in Finnish texts compared to languages like English.
- First Seen Age (FSA): TCBLex introduces the FSA metric for Finnish, identifying the earliest age at which a word or lemma appears in the corpus. This serves as a valuable proxy for the "age of acquisition."
- Syntactic Evolution: The data reveals a shift in narrative structure as reading age increases; the proportion of nouns decreases while the use of pronouns and auxiliary verbs rises, suggesting a move toward more internal and complex dialogue.
Conclusion and Future Directions
TCBLex represents a major milestone for Finnish psycholinguistics and education. While the current version is highly accurate, future work will focus on expanding the corpus to include more textbooks and extending analysis to syntactic complexity at the sentence level.
Availability: TCBLex is freely available for research purposes at Zenodo.
Summary updated