Yellowhammer dialectsENRO
Map of Europe with 1,178 classified yellowhammer recordings; where the licence allows, a cell shows the spectrogram of one bird’s song ending, coloured by its type; the others are dots.
Spectrogram of one yellowhammer song: a run of short notes followed by a high note and a long level note.the runthe ending
high note first
falling whistle last
higher note last
lower note last

Drawing the map…

    1. One bird · 1,178 recordings · 28 countries

      Yellowhammer song dialects: a map of Europe drawn from the birds’ own song endings

      Birds have accents. The yellowhammer (Emberiza citrinella), a small yellow bird of hedges and field edges, sings one short song and ends it differently from region to region. Each tile on this map is how one real bird ends its song. Tap a tile to hear it.

      Find the recordings nearest your townor scroll for the story

    2. The song

      A run of quick notes, then a long one#

      The yellowhammer is a sparrow-sized bunting of hedges and field edges; the male has a bright yellow head. A song lasts about 2.4 seconds.

      This is one real song. Time runs from left to right and pitch from bottom to top. Left of the dashed line is the run of quick, even notes. Right of it is the ending, and the ending is the part that changes from one region to the next.

    3. The endings

      Four ways to end the same song#

      Nearly every ending has one long, level note at about 5 kHz. The difference is what comes with it: a high note before it, a whistle that falls after it, a note that sits higher or climbs after it, or a second note lower down. A few birds do two of these.

      Ornithologists write these with letters (XB, BC, BE, BhBl). Here each is named for what is heard. Press one to listen to a bird that sings it.

    4. North Sea to Scandinavia

      From the Netherlands to Norway, the high note comes first#

      In the Netherlands, Belgium, Germany and Switzerland, 271 of 298 classified birds sing a high note just before the long one. Sweden and Norway sing the same way: 41 of 43. Denmark, between them, has only 7 classified birds, and 5 of them end on the falling whistle.

      “Bird” here means recordings counted once per recordist, per 10 km and per ending, so that one keen recordist in one wood cannot colour a region. These are not verified individual birds.

    5. The Baltic

      Across the Baltic, the song ends another way#

      Sweden is amber. Across the Gulf of Bothnia and the Baltic, Finland and Estonia are blue: 36 of 40 classified birds there end on a falling whistle, and the high note before the long one is almost gone.

      The recordings cannot show that the sea is the cause. They show that two regions meet there.

    6. Poland and Austria

      In Poland two birds in three end on a higher note#

      Between the amber west and the blue east lies a third ending: after the long note comes a second one, higher, level or climbing. 47 of the 69 classified Polish birds sing it; the 69 come from 28 recordists, one of whom recorded 22. Most of the others end on the falling whistle.

      Austria is divided: of 27 classified birds, 11 end on the higher note and 13 on the falling whistle. Czechia, between them, has a little of everything; the Czech researchers who surveyed it call it a mosaic.

    7. East of 20° E

      From Estonia to Ukraine and Russia: the falling whistle#

      In Estonia, Finland, Russia and Ukraine together, 90 of 104 classified birds end on the falling whistle.

      The high note that opens the ending in the west thins out between 15° and 20° E and is almost absent further east. It opens 75% of classified endings between 10° and 15° E, 28% between 15° and 20° E, 11% between 20° and 25° E, and 2 of 80 between 25° and 40° E.

    8. Britain

      Britain: three patterns on one island, and a note this map misses#

      In southern England the high note comes first, as across the Channel: 25 of 28 classified birds south of 52° N. Between 52° and 54° N the tiles turn blue: 14 of 19 end on the falling whistle, and where the researchers’ own map has birds nearby (7 cases) it says the same. North of 54° N the picture is mixed again: 11 of 20 with the high note first.

      One place shows what this map cannot see. On the Yorkshire Wolds one recordist has uploaded 193 recordings from a single small area. The researchers’ citizen-science map writes the birds there with a very short high note first: 17 of its 39 points as XsBC (short high note, long note, falling whistle) and 15 as XsB, without the whistle. The labelling used here called 147 of the 193 recordings a plain falling whistle. Where the same recording could be matched in both sets, 9 of 10 songs written XsBC appear here without the short note.

      So a blue tile says that the song ends on a falling whistle. It cannot say whether a very short high note comes first.

    9. The test

      Neighbours sing alike, for hundreds of kilometres#

      92%of pairs of birds less than 25 km apart share an ending; by chance, 42%

      Take two classified birds recorded by different people less than 25 km apart. They share an ending in 92% of such pairs. Shuffle the endings among the same places and it is 42%. Most of those close pairs are Dutch, and the Netherlands sings almost one ending; without it the figure is 85%.

      The likeness fades with distance and falls to the shuffled level at about 700 km. Beyond that, two birds share an ending less often than shuffled ones do, as a few large regions would produce.

      A guess made from the nearest classified bird recorded by someone else is right for 81% of birds; always guessing the commonest ending would be right for 57%.

    10. What is not on the map

      Most recordings stayed grey#

      The archives hold 6,743 located recordings of the species. 6,515 could be opened, and in 5,186 a song was found whose ending could be drawn. Two AI models, working separately and without knowing the place, named the same complete ending for 1,194; 1,178 of those are in Europe and on this map.

      The others with a drawn ending are the grey dots: a long note with nothing readable after it, a faint recording, a bird that stopped short. Grey means “not classified”. It does not mean another dialect.

      The two models were first tried on 97 recordings for which the researchers have published a dialect. Where both named the same complete ending they matched the published one in 49 of 51. The trial was nearly all xeno-canto recordings; for phone recordings the only check is the comparison by place in the method below. The models’ usual mistake is to lose a faint element, which leaves the bird grey and thins the map where recordings are poor, as in south-eastern Europe. The mistake that matters more is the one in the Britain chapter: a very short high note is dropped and the bird gets another colour. The trial held no such songs and could not measure it.

    11. What was known

      A century of study, and what the open recordings add#

      The yellowhammer’s dialects have been studied since the 1920s. In 2015 Tereza Petrusková and colleagues pooled the published studies with 204 recordings from online sources and sound archives and concluded that the continent’s pattern does “not support the existence of broad, geographically distinct dialect groups (eastern vs. western)”.

      The review’s own map sorts the dialects into an “X” group, with the high note before the long one, and a “B” group, without it. A rule for testing its conclusion on the open recordings was written down before they were labelled. It would count as support if endings with and without the high note each made up at least 15% of birds west of 5° E, between 5° and 20° E and east of 20° E, and as a contradiction if either fell under 5% in a band of at least 50 birds. East of 20° E the high note came out at 7% (95% interval 2 to 14%): neither. It stays above 5% only because 7 of the 12 such birds are in northern Sweden and Åland, just east of the line.

      The slope is not new. The review’s own list has 1 such ending among 34 recordings between 20° and 70° E, and the researchers’ map, with thousands of recordings classified by hand, 6 among 1,987. No ending belongs to one half of the continent alone: the falling whistle is sung in northern Spain, by birds from 5 recordists, and in Russia, though this map cannot tell whether a very short high note comes before it in Spain.

      What the open archives add is a second reading, made without the researchers’ labels, of recordings that anyone can hear, 88% of them made in 2015 or later. They also add ground: of the 381 squares of 50 km in which this map has a classified bird, the researchers’ map has no classified point in 172. Whether this reading of the archives is right is for the researchers to say.

    12. Romania

      Romania: eight recordings, none clear enough#

      The archives searched here hold 8 yellowhammer recordings from Romania, the oldest from 2022. In 3 of them the long note is there and nothing after it can be read; in the others no ending could be made out. Neither model could name a complete ending for any.

      The researchers’ own map has 3 more, all from one June day in 2014 in the Banat mountains: one classified, with a lower note last, and 2 with only the long note.

      The neighbours are the only hint. In Ukraine 15 of 18 classified birds end on the falling whistle. Hungary has 5 classified birds and three endings among them. Bulgaria has 15 recordings with a song and none that could be classified. East of 25° E the high note before the long one is rare, though the researchers’ map has one such bird at Dulovo, in north-eastern Bulgaria. What Romanian yellowhammers sing cannot be read from the recordings in these archives.

    13. What do yellowhammers sing near you?

      Type the name of a town. The map shows the three nearest classified recordings by different people, how far away they are and which ending they carry.

      No town chosen. The map shows all 1,178 classified recordings.

      Tap any tile or coloured dot to hear that bird; grey dots are recordings that could not be classified. Drag the map with a mouse; the + and − buttons zoom. Press a coloured label to see one ending only. Each tile stands for the classified recordings that fall under it on the screen, and shows the clearest one with the commonest ending. A corner in a second colour marks a tile where no ending reaches two thirds of them; tap the tile to see the count.

    How the map was made

    Every sound recording of Emberiza citrinella with coordinates in the Global Biodiversity Information Facility on 11 October 2026 was listed: 6,743 recordings, from xeno-canto, iNaturalist, observation.org and 5 smaller sources. 6,515 could be downloaded; the first two minutes of each were used.

    A program finds the songs (a run of at least five evenly spaced notes), lines up the endings of all the songs in a recording and averages them into one picture. Averaging removes other birds and wind and keeps what the bird repeats. The same program tried to read the endings and could not: on recordings with a published dialect it gave the published label about one time in three. It labels nothing on this page.

    The pictures were then labelled by sight, as the published studies do. Two AI models from different makers, Claude (Anthropic) and Codex (OpenAI), each saw every picture on numbered sheets with no place, date or name, and chose among the published types. A recording is classified only if both named the same complete ending. For 66 recordings they named two different complete endings; those stay grey. No ornithologist has checked the labels.

    Of the 1,194 classified recordings, 1,178 are on the map. The other 16 are from New Zealand, where the bird was introduced from Britain in the nineteenth century, and from Asia east of 70° E.

    What the types are

    The letters are those of the nomenclature introduced by Poul Hansen in 1985 and used by the Czech researchers. Their meaning was read here from the example sounds on the researchers’ project site: B is the long note, X a high note before it, and C, E, Bh and Bl name what follows.

    Two reductions were needed. The first concerns the family with a higher note last. The researchers’ 2023 study found that its three published types (BE, BD and BlBh) are separated by arbitrary frequency thresholds and that two groups describe them better than three, at least in Central Europe; in the data published with it, 59 of the 69 points first written BD are moved to BlBh and the rest to BE or left as transitional. The two models were tried on 60 single songs from that study: 19 BE, 18 BlBh and 23 BD. They named the same complete ending for 46. All 25 of the BE and BlBh songs among those were read as published. Of the 21 BD songs among them, the in-between type, the models read 15 as BE, while the study moves 18 of those 21 to BlBh. The models tell the two ends apart but draw the line through the middle in another place than the researchers do, so the family is shown here as one type, “higher note last”. Second, a very short X is not seen reliably, so XsBC appears as BC, and the “high note first” type is in practice the longer X.

    Where the sounds and pictures come from

    Each recording keeps its recordist’s licence, shown with it. A clip is an excerpt of about three seconds, with low rumble filtered out and the level evened; a picture is the average of the song endings in one recording. Both are adaptations, offered under the same licence as their source. Recordings under a no-derivatives licence (464 on the map) appear as a dot with a link to the original; no picture or clip was made from them.

    The numbers behind the chapters

    In the tables a bird is counted once per recordist, per 10 km and per ending; a recording uploaded to two archives counts once. The bands stop at 70° E.

    Share of classified birds with a high note before the long one, by longitude
    BandBirdsHigh note first95% intervalxeno-canto only: birdsHigh note first
    west of 5° E13567%54–80%9066%
    5° to 20° E43875%59–86%17772%
    20° to 70° E1727%2–14%7710%

    The same share in steps of ten degrees

    Bands with fewer than 30 birds are left out.

    High note first, by ten degrees of longitude
    BandBirdsRecordistsHigh note first
    10°–0° W857165%
    0°–10° E29422087%
    10°–20° E19414257%
    20°–30° E123648%
    30°–40° E39343%

    How alike two birds are, by distance

    Pairs of classified birds recorded by different people. “Shuffled” is the mean of 999 random reassignments of the endings among the same places.

    Pairs of birds that share an ending
    Distance apartPairsSame endingShuffled
    0–25 km1,50592%42%
    25–50 km2,20286%42%
    50–100 km5,46779%42%
    100–200 km10,04373%42%
    200–400 km26,80367%42%
    400–800 km74,93651%41%
    800–1,600 km107,25330%41%
    1,600+ km48,28023%41%

    Checks that could have cancelled the map

    Published labels. The trial set was 97 recordings: 11 from the 2015 review’s list and 86 matched by date and place, within 300 m, to points on the researchers’ map. Of the 92 with a published complete ending, the two models named the same complete ending for 51 and matched the published one for 49.

    The researchers’ map, by place. Outside Britain, 463 classified recordings have a classified point of the researchers’ map within 10 km; 429 carry the ending that is commonest among those points (93%). For iNaturalist recordings, most of them made with a phone, it is 33 of 40. Some of these pairs are the same recording in both sets, so this is partly a comparison of two labels for one file.

    Czechia. Of the Czech recordings classified here, 19 lie within 5 km of a male classified in the Czech survey; 17 carry the same ending. Several are the same place counted more than once.

    Phones against microphones. A faint high note is easier to lose on a phone. In Germany the high note opens 88% of classified iNaturalist birds and 95% of xeno-canto ones; in Poland 22% and 7%; in Britain 59% and 60%; in France 93% and 65%. The archives differ by 1 to 28 points, in no steady direction. The bands either side of 20° E differ by about seventy.

    Location and crowding. Leaving out recordings whose stated position uncertainty is over 1 km (in practice most of the Dutch ones), pairs less than 25 km apart share an ending in 84% of cases; leaving out the Netherlands altogether, 85%. Half the birds, all of xeno-canto, state no uncertainty.

    One recording per square. Keeping a single recording in each 10 km square leaves 634 recordings. Pairs of them made by different people less than 25 km apart share an ending in 89% of cases, against 40% when shuffled; without the Netherlands, 90% against 37%.

    Each model alone. Asking two models to agree could itself select certain recordings. Taking each model on its own, with every bird to which it gave a complete ending, the pattern stays: in the Netherlands, Belgium, Germany and Switzerland the high note comes first in 311 of 343 birds by Claude alone and 299 of 340 by Codex alone; east of 20° E, in 17 of 193 and 14 of 205.

    What got classified, by region. Of the recordings with a drawn ending, 16% were classified between 10° W and 0°, 29% between 0° and 10° E, 25% between 10° and 20° E, 30% between 20° and 30° E and 18% between 30° and 40° E. The share changes little at 20° E, where the endings change (25% to 30%). It is lowest in the west, where most recordings are British, and east of 30° E. The grey dots on the map are the recordings left unclassified.

    Measured, not quoted. The length of a song is the median of 151 classified recordings here. The pitch of the long note is the median of 430 Czech songs measured by Diblíková and colleagues (2023).

    Without labels. Sorting the 702 shipped pictures of the three commonest types into four groups by shape alone, with no labels, puts 71% of them in a group dominated by their own type; all in one group would give 57%. About 71% of the high-note-first pictures fall into groups that hold almost nothing else; the rest, and the two kinds of tail, are not separated by shape alone.

    What remains unchecked. The labels have not been reviewed by the researchers who built the nomenclature. A very short high note is missed, as the Britain chapter shows. Spain, Ukraine, Estonia and parts of Russia rest on about ten recordists each or fewer. South-eastern Europe is almost empty. About 19% of iNaturalist recordings with a drawn ending could be classified and 24% of xeno-canto ones, so the map follows where good recordings exist.

    Sources

    The people who made the recordings

    The 714 classified recordings whose pictures and sounds appear on the map, by recordist, each with its licence. Each link opens the original.

    374 more recordists