Measure of Textual Lexical Diversity (MTLD)
Mean MTLD per year, computed after McCarthy & Jarvis (2010). Only songs with at least 100 tokens are included, which excludes 675 of 15,893 songs but only 0.9 % of all tokens. Select an archive below to narrow the view. For window-based alternatives, see Type-Token Ratio (STTR, MATTR) →
Linguistic Insights
MTLD confirms the window-based picture
MTLD ranks the archives almost exactly as STTR and MATTR do (rank correlation 0.97 across the 14 archives; 0.86 with MATTR per song). HipHop Songs (MTLD ≈ 128) sits in the top group with Fettes Brot, Hannes Wader and Konstantin Wecker, well above Chart Songs (≈ 75), NDW Songs (≈ 79) and GDR Songs (≈ 81). MTLD measures how long a running text sustains a diversity threshold; because it is not capped at 1, it spreads the most varied archives further apart than a fixed-window ratio does (see Type-Token Ratio →).
Read small archives and sparse years cautiously
Per-song MTLD ranges from under 10 to over 1000, so yearly means in small archives can swing on a few outlier songs, and the y-axis rescales to whatever is in view. Sparse years — fewer than three songs — appear as isolated dots rather than joined to the line. For an artist archive, a single productive year — a new album with unusually varied or unusually formulaic language — can move the archive mean substantially.
Mean MTLD per year · songs with ≥ 100 tokens · factor threshold 0.720 · dots = fewer than 3 songs in that year
These analyses are based on corpus data as of August 31, 2026.