The project and its needs
The FethiGedikli.com Dictionary is an academic dictionary of words found in legal and literary texts, dictionary and language studies, regional usage, newspapers, television and digital sources, presented with their semantic relations, examples, sources, people, etymology, region and dates.
Because entries are written as free text, person, source and date information had to be extracted, relations between words built, and all of it turned into searchable indexes.
Automatic semantic relations through rules
Relations between words are inferred by rules: for example, a definition such as "a kind of X" gives a broader term, the -msı suffix gives a similarity link, and spellings that differ only by a circumflex give a confusable or spelling-variant link. Every link carries the rule that produced it and a confidence level from 1 to 3; links added by hand survive automatic rebuilds.
- Synonyms, near-synonyms, antonyms; similar; differences; confusables
- Broader and narrower terms, direct and indirect relations
- "When there is a Turkish word" links
- Clusters of 3–9 and larger word groups derived from the relation graph
Indexes extracted from the text
People, years and sources are extracted automatically from definitions and examples; newspaper abbreviations are expanded and sources are grouped by type (newspaper, television, social media, official, academic).
- Indexes by field, part of speech, origin, affix, person, source, date and region
- 18 parts of speech and 18 languages of origin
- 1,395 person cards, 425 sources, 44 regions
- Indexes rebuilt automatically without downtime
Search and reader features
- Finding the headword from inflected or suffixed searches
- 11 fields × 4 match modes (contains, starts with, ends with, exact)
- Filters for relation type, cluster size, "has example" and "has source"
- A relation panel that shows the evidence rule, and two-way "see also"
- Word of the day, latest additions and most searched
- Results from TDK, Lugatim and FaRkLaR.net on the same page
Automation and data quality
A task that runs every minute moves quotations from definitions into the example field, source markers into the source field and URLs into the web field, and flags duplicate headwords without merging them. Searches with no result are logged as "words to add" hints.
What this means for your project
If your dictionary is written as free text, we can use the same approach to extract person, source and relation indexes and quickly turn it into a searchable reference.