In Search of the Weirdest Language

This is a replay of Tyler Schnoebelen's experiment on finding the most typologically unusual language, set up with somewhat different conditions. Like him, I started with the WALS table, but decided to focus on a fragment with no gaps.

The Experiment

I first excised all features with non-A codes (as they tend to be elaborations of A-coded ones) and all that have ‘Other’ among their values (because such a value can make languages appear similar even if they aren't). Of what was left I chose 200 languages and 20 features, such that everything has a value for everything. (I resisted the temptation to fill any gaps or correct any entries in the WALS table, although it wasn't easy. I knew that if I began doing that, there'd be no end.) That is, among the 218 languages that have values for all 20 features I chose those that either are in the WALS 200-language sample or belong to families with no more than seven representatives (actually, only Amharic reaches this limit).

Languages:

//Ani, Abipón, Abkhaz, Achumawi, Ainu, Alamblak, Amele, Amharic, Apurinã, Arabic (Egyptian), Araona, Arapesh (Mountain), Armenian (Eastern), Asmat, Awa Pit, Aymara, Bagirmi, Bambara, Basque, Batak (Karo), Bawm, Beja, Bribri, Burmese, Burushaski, Cahuilla, Campa (Axininca), Canela-Krahô, Carib, Cayuvava, Chamorro, Chinantec (Lealao), Chukchi, Comanche, Coos (Hanis), Cree (Plains), Cubeo, Daga, Dagur, Dani (Lower Grand Valley), Diola-Fogny, Drehu, Dumo, Ekari, English, Epena Pedee, Evenki, Ewe, Fijian, Finnish, French, Fur, Garo, Georgian, German, Gooniyandi, Grebo, Greek (Modern), Greenlandic (West), Guaraní, Haida, Hamtai, Hausa, Hindi, Hixkaryana, Hmong Njua, Hungarian, Hunzib, Ika, Imonda, Indonesian, Iraqw, Jakaltek, Japanese, Ju|'hoan, Kannada, Kanuri, Karok, Kawaiisu, Kayah Li (Eastern), Kayardild, Kera, Ket, Kewa, Khalkha, Khasi, Khmer, Khmu', Khoekhoe, Kilivila, Kiowa, Kiribati, Koasati, Kobon, Korean, Koromfe, Koyraboro Senni, Krongo, Kunama, Kutenai, Ladakhi, Lakhota, Lango, Latvian, Lavukaleve, Lepcha, Lezgian, Luvale, Maba, Maidu (Northeast), Malagasy, Mandarin, Mangarrayi, Maori, Mapudungun, Maranungku, Maricopa, Marind, Martuthunira, Maung, Maybrat, Meithei, Miwok (Southern Sierra), Mixtec (Chalcatongo), Mundari, Murle, Nahuatl (Tetelcingo), Nasioi, Navajo, Ndyuka, Nenets, Nez Perce, Ngiti, Nivkh, Nkore-Kiga, Nunggubuyu, Oneida, Oromo (Harar), Otomí (Mezquital), Paamese, Paiwan, Passamaquoddy-Maliseet, Paumarí, Persian, Pirahã, Pitjantjatjara, Pomo (Southeastern), Purépecha, Qawasqar, Quileute, Rama, Rapanui, Russian, Sango, Sanuma, Semelai, Sentani, Shipibo-Konibo, Slave, Spanish, Squamish, Suena, Supyire, Swahili, Taba, Tagalog, Thai, Tiwi, Tlingit, Tonkawa, Trumai, Tsimshian (Coast), Tukang Besi, Tunica, Turkish, Una, Ungarinjin, Urubú-Kaapor, Usan, Vietnamese, Wambaya, Wappo, Warao, Wardaman, Wari', West Makian, Wichí, Wichita, Wintu, Yagua, Yaqui, Yidiny, Yimas, Yoruba, Yuchi, Yukaghir (Kolyma), Yurok, Zoque (Copainalá), Zulu, Zuni.

Of these 93 are in the WALS 100-language sample, 90 are in the 200-language sample but not in the 100-language one, and 17 are in neither, but happen to be well-documented.

Features:

  1. 1A Consonant Inventories.
  2. 2A Vowel Quality Inventories.
  3. 3A Consonant-Vowel Ratio.
  4. 4A Voicing in Plosives and Fricatives.
  5. 6A Uvular Consonants.
  6. 7A Glottalized Consonants.
  7. 8A Lateral Consonants.
  8. 11A Front Rounded Vowels.
  9. 18A Absence of Common Consonants.
  10. 19A Presence of Uncommon Consonants.
  11. 44A Gender Distinctions in Independent Personal Pronouns.
  12. 48A Person Marking on Adpositions.
  13. 82A Order of Subject and Verb.
  14. 83A Order of Object and Verb.
  15. 100A Alignment of Verbal Person Marking.
  16. 102A Verbal Person Marking.
  17. 103A Third Person Zero of Verbal Person Marking.
  18. 104A Order of Person Markers on the Verb.
  19. 107A Passive Constructions.
  20. 143A Order of Negative Morpheme and Verb.

The Results

Then I calculated the blandness of every value of the table as the number of times it appears for that value (for example, English has no uvular consonants, and it's one of 162 languages that have none, so that's a blandness of 162/200=0.81). Every language's overall blandness is the harmonic mean of its blandness values for all features.

The languages again, in increasing order of weirdness (the equals sign means a tie between the two blandest languages, because mirabile dictu they share all values despite having nothing in common):

Koyraboro Senni = Yaqui, Ndyuka, Una, Bambara, Khasi, Meithei, Bribri, Thai, West Makian, Chinantec (Lealao), Suena, Bawm, Daga, Mundari, Purépecha, Evenki, Alamblak, Kilivila, Khalkha, Usan, Tagalog, Comanche, Imonda, Hindi, Fur, Kobon, Russian, Shipibo-Konibo, Kannada, Garo, Araona, Kunama, Koromfe, Taba, Diola-Fogny, Kayah Li (Eastern), Maung, Korean, Cahuilla, Tonkawa, Kanuri, Achumawi, Paamese, Indonesian, Maori, Tukang Besi, Dani (Lower Grand Valley), Marind, Tiwi, Mixtec (Chalcatongo), Yidiny, Malagasy, Sanuma, Beja, Ika, Sentani, Asmat, Nkore-Kiga, Supyire, Otomí (Mezquital), Qawasqar, Martuthunira, Maba, Wappo, English, Miwok (Southern Sierra), Gooniyandi, Basque, Fijian, Zuni, Rapanui, Greek (Modern), Bagirmi, Mangarrayi, Dumo, Tunica, Apurinã, Burmese, Hmong Njua, Maybrat, Pomo (Southeastern), Campa (Axininca), Batak (Karo), Ungarinjin, Ainu, Chamorro, Nahuatl (Tetelcingo), Amharic, Abipón, Swahili, Arapesh (Mountain), Kayardild, Latvian, Maranungku, Drehu, Kiribati, Yoruba, Georgian, Kewa, Persian, Greenlandic (West), Vietnamese, Mapudungun, Luvale, Krongo, Khmer, Murle, Nivkh, Turkish, Wambaya, Grebo, Ewe, Koasati, Yurok, Finnish, Amele, Maidu (Northeast), Wardaman, Urubú-Kaapor, Hungarian, Karok, Yagua, Warao, Paiwan, Guaraní, Hixkaryana, Spanish, Lango, Hamtai, Ket, Slave, Maricopa, Burushaski, Kera, Haida, Yimas, Aymara, Carib, Epena Pedee, Cree (Plains), Lakhota, Passamaquoddy-Maliseet, Sango, Japanese, Canela-Krahô, Dagur, Yuchi, Navajo, Jakaltek, Wintu, Pirahã, Kawaiisu, Rama, Nez Perce, Cayuvava, Coos (Hanis), Hausa, Chukchi, Ngiti, Hunzib, Ju|'hoan, Lavukaleve, Trumai, Mandarin, Oromo (Harar), Arabic (Egyptian), Tsimshian (Coast), Cubeo, Nunggubuyu, Pitjantjatjara, Khoekhoe, Quileute, Kutenai, Lezgian, Oneida, Ladakhi, Wichí, French, Squamish, Lepcha, Zoque (Copainalá), Yukaghir (Kolyma), Wichita, Nasioi, Semelai, Ekari, Zulu, Paumarí, //Ani, Awa Pit, Wari', Tlingit, Armenian (Eastern), Kiowa, German, Nenets, Khmu', Abkhaz, Iraqw
On the map the 130 blandest languages are marked by diamonds and the 70 weirdest ones by circles, with the colours progressing from white (the two blandest ones) to red-orange-yellow-green-cyan-blue-black (the weirdest one). One may note that almost all languages in India are in the blandest one-sixth, as are half of the languages of Papua New Guinea; that in Africa the blandest languages are all on almost the same latitude, but on different longitudes from far west to far east; and that in the Americas most languages are relatively weird, the bland ones being few and scattered wide, but not very far north or south.

Overall there is a −0.308492644 correlation between a language's blandness and its distance from the 13½th parallel north. Of the languages up to 26 degrees north and south of this parallel, 68 are at least as bland as Batak (Karo) and 62 are at least as weird as Ungarinjin; beyond these latitudes the corresponding numbers are 16 (bland) and 54 (weird).

Conlangs

If Esperanto were included, it would end up in the 17th place, between Purépecha and Evenki (which happens to be the blandest Nostratic language). So it does fulfil its purpose of being less weird than all Indo-European languages (well, at least the ones in my table).

If Volapük were included, it would end up in the 97th place, between Maranungku and Yoruba. It would still be in the blander half, but only just.

If Klingon were included, it would be last but one (200th of 201), between Abkhaz and Iraqw. Thus it is very weird indeed, but not much weirder than all Terran languages — not even weirder than all of them. It even shares the values of six features with English (a.k.a. Federation Standard), from which it tries to be as dissimilar as possible (1A an average-size consonant inventory, 4A voicing in both plosives and affricates, 7A no glottalised consonants, 11 no front rounded vowels, 11A all common types of consonants present, 48A no person marking on adpositions).

The Other 18 Languages

The 18 languages that also have values for the 20 features but didn't make it into the experiment, because they aren't in the WALS 200-language sample and belong to well-represented families, are from blandest to weirdest:
Koyra Chiini, Lahu, Adzera, Polish, Nandi, Berta, Kashmiri, Komo, Zande, Mumuye, Tigak, Albanian, Yapese, Atayal, Kurdish (Central), Malakmalak, Ijo (Kolokuma), Doyayo.
In the rank ordering Koyra Chiini fits in the third place, between its sister Koyraboro Senni (as well as Yaqui) and Ndyuka. (The only difference is that Koyra Chiini is VO rather than OV.) Lahu is between Dani (Lower Grand Valley) and Marind; Adzera, between Qawasqar and Martuthunira. The place of Polish is between Fijian and Zuni. (It is significantly more weird than Russian because it has a large consonant set, as opposed to a moderate one, and especially because it's undecided between SV and VS.) Of the other Indo-European languages, Kashmiri is between Maranungku and Drehu (in the bland half, but close to the border); Albanian, between Tsimshian (Coast) and Cubeo; Kurdish (Central), between Kutenai and Lezgian. Ijo (Kolokuma) fits between Semelai and Ekari. But the weirdest one is Doyayo, which ranks in the last place but one, between Abkhaz and Iraqw.

Weirdest Features of Some Languages

The weirdest things about both Koyraboro Senni and Yaqui are that they have no verbal person marking (this affects two features: 100A and 102A). There are 48 languages among the 200 that are like that, but with respect to everything else, the two blandest languages are in even more numerous company.

The weirdest thing about Esperanto is that its personal pronouns distinguish gender, albeit in the 3rd person singular only (44A). This value is shared by 30 languages.

The weirdest thing about Hindi is its large consonant inventory (1A), 34 segments or more (UPSID lists 38).

The weirdest thing about Russian is its high consonant-vowel ratio (3A), above 6.5 (UPSID finds 33 consonants and 5 vowels, yielding a ratio of 6.6).

The weirdest thing about Indonesian is that only the patient argument is marked in the verb (102A).

The weirdest thing about English are its interdental fricatives (19A).

The weirdest thing about Volapük are its front rounded vowels, both mid and high (11A).

The weirdest thing about Mandarin is its high front rounded vowel (also 11A).

The weirdest thing about German is that its negative particle can both precede and follow the verb (143A).

The weirdest thing about Abkhaz is that its negative affix can both precede and follow the root (143A).

The weirdest thing about Klingon is the tripartite alignment of its verbal person marking (100A), something unseen in my sample.

The weirdest thing about Doyayo is that it has optional triple and obligatory double negation (143A), which is also unseen in the sample.

The weirdest thing about Iraqw is that its personal pronouns distinguish gender, but not in the 3rd person (44A), also a feature that no other language in the sample has.

There are only two other languages that are unique in some respect: Nenets has both interdental fricatives and pharyngeals (19A), and Khmu' has both implosives and glottalised resonants (7A).

The most blandest language ever

Doesn't exist; but it would possess the following features (that is, feature values), of which every language in my sample shares between 6 (Quileute) and 16 (Una, Khasi, Bawm, Maung and Indonesian — note that these are all of different families):
  1. An average consonant inventory (19 to 25 segments).
  2. An average number of vowel qualities (5 or 6).
  3. An average consonant to vowel ratio (no less than 2.75 but less than 4.5).
    (Which is to say, 19 to 22 consonants and 5 vowels or 19 to 25 consonants and 6 vowels.)
  4. No voicing contrast in plosives and fricatives.
  5. No uvular consonants.
  6. No glottalised consonants.
  7. /l/ but no obstruent literals.
  8. No front rounded vowels.
  9. All kinds of common consonants (bilabials, fricatives and nasals).
  10. No uncommon consonants (clicks, labial-velars, pharyngeals or interdental fricatives)
  11. No gender distinctions in independent personal pronouns.
  12. No person marking on adpositions.
  13. A subject-verb order.
  14. An object-verb order.
    (But it needn't be SOV; it may be OSV. My count won't know.)
  15. Accusative alignment of verbal person marking.
  16. Both the A and P arguments marked on the verb.
  17. No zero realisation of 3rd person on the verb.
  18. Neither A nor P, or only one of them, marked on the verb.
  19. No passive constructions.
  20. Negation expressed by a preverbal particle.
Note that there is a conflict between the 16th and the 18th value: features 102A and 104A aren't independent. Of these two values ‘Both the A and P arguments marked on the verb’ is the more common one. If the two are combined into a single feature, ‘A precedes P’ wins over ‘No person marking’, though very slightly (49 to 48).

Most and Least Similar Languages

Even if all 218 fully documented languages are considered, and Esperanto, Volapük and Klingon are added, Koyraboro Senni and Yaqui remain the only two languages to share the values of all 20 features. They differ from Koyra Chiini in only one place, as do Chinantec (Lealao) and Chamorro, Daga and Usan, and Evenki and Fur. The two languages that differ most from the ones from which they differ least are Abkhaz and Paumarí: the former has 9 values in common with Burushaski and Tlingit, the latter with Krongo and Russian.

No two languages differ in all 20 values, but Nunggubuyu differs from Iraqw in 19, as do Wambaya and Oneida from //Ani, and Burmese and Hunzib from Wari'. Nine languages (Basque, Comanche, Daga, Hindi, Kanuri, Kunama, Mundari, Purépecha, Una) differ least (in 14 values) from the ones from which they differ most.

Indonesian differs least (in 4 values) from Chinantec (Lealao), Dagur, Hindi, Khasi and Russian, and most (in 15) from Tlingit.

English differs least (in 4 values) from French, Russian and Volapük, and most (in 17) from Tlingit.

Esperanto differs least (in 3 values) from Mandarin, and most (in 16) from Tlingit.

Volapük differs least (in 4 values) from Dagur, English, Finnish and French, and most (in 17) from Tlingit again.

Tlingit differs least (in 8 values) from Aymara, Cahuilla, Navajo and Wichita, and most (in 18) from Cubeo, Drehu, French, German and Latvian.

Abkhaz differs least (in 9 values) from Burushaski and Tlingit, and most (in 18) from Maori.

Klingon differs least (in 8 values) from Kunama, and most (in 16) from Finnish, Kayardild and Martuthunira.

To be sure, a simple count of shared feature values, with no weights assigned to them, is a very rough measure of the similarity between languages. If we use it to divide all 200 sample languages into two classes, so as to maximise the number of token differences between languages of different classes and to minimise the number of differences within each class, the languages will split almost by features 102A and 104A (that is, by whether the verb expresses both the A and P arguments or not). (The three exceptions are Kanuri, Kunama and Maba, all of which do express both A and P in the verb but otherwise share slightly more values with languages of the other class.) This is an effect of the correlation between features 102A and 104A (as well as 100A and 103A).

Slashing Correlated Features

It was so hard to find 20 features that have values in WALS for 200 languages, it wouldn't do to discard any altogether. But we can merge 102A and 104A, because they're in complementary distribution: any language has either ‘5 Both the A and P arguments’ for the former or ‘1 A and P do not or do not both occur on the verb’ for the latter but not both. Also they correlate with 100A and 103A: any language that has ‘1 Neutral’ for 100A has ‘1 No person marking’ for 102A (and vice versa), has (with one exception) ‘1 No person marking’ for 103A, and has ‘1 A and P do not or do not both occur on the verb’ for 104A. So when counting how many values two languages have in common, let's weigh 100A, 102A*104A and 103A by ¼, ½ and ¼ respectively.

If we use these weights when dividing all 200 sample languages into two classes as said above, the split turns out to be mainly by feature 83A Order of Object and Verb: all languages with VO or no dominant order are in one class (97 members), all OV languages are in the other (103 members). There are three exceptions: Wichita (OV) is in the first class and Karok and Trumai (no dominant order) are in the second. Unsurprisingly, this division correlates with 82A Order of Subject and Verb, because all languages in the second class are SV, except for Hixkaryana and Cubeo. And also, less strongly, with 143A Order of Negative Morpheme and Verb: the first class contains more, and the second less, of their fair shares of negative preverbal particles, and with negative suffixes it's the other way around.

Distance as Number of Mismatches

Or what's going to happen if we count how far each language is removed from all 200 sample languages, meaning by that in how many values it differs? By summing up the distances we'll get a measure of the language's weirdness. It will be a rough measure unless (or until) weights are added, but still.

From blandest to weirdest, they go (non-sample languages in italics):

Indonesian, Una, Bawm, West Makian, Hindi = Koyraboro Senni = Yaqui, Khasi, Koyra Chiini, Basque, Ndyuka, Maung, Purépecha, Daga = Kunama, Meithei, Suena, Evenki = Mundari, Kanuri, Comanche, Bambara, Beja, Fur = Taba, Bribri, Thai = Tonkawa, Marind, Paamese = Russian, Miwok (Southern Sierra), Kobon, Chinantec (Lealao) = Usan, Imonda, Abipón, Kannada, Tiwi, Amele, Maba, Achumawi, Tukang Besi, Dagur, Alamblak, Esperanto, Khalkha = Ladakhi, Dani (Lower Grand Valley), Diola-Fogny = Rama, Kewa, Batak (Karo), Maidu (Northeast) = Sentani, Kilivila, Lavukaleve, Kayah Li (Eastern), Araona, Komo, Asmat, Cahuilla = Koromfe, Warao, Chamorro = Mangarrayi = Shipibo-Konibo = Tagalog, Dumo, Garo, Lahu, Gooniyandi, Luvale, Supyire, Wappo, Hmong Njua = Karok, Tigak, Zoque (Copainalá), Urubú-Kaapor, Bagirmi = Tunica, Polish, Nkore-Kiga, Fijian, Chukchi, Nenets = Yoruba, Ainu, Ika, Hungarian, Sanuma, Mixtec (Chalcatongo) = Yurok, Turkish, Adzera = Guaraní, Ewe = Maori, Yukaghir (Kolyma), Canela-Krahô, Nahuatl (Tetelcingo), Korean, English = Qawasqar, Nandi, Ekari, Swahili, Vietnamese, Otomí (Mezquital), Ungarinjin, Volapük, Greek (Modern) = Kashmiri, Yidiny, Mapudungun, Lango, Kera, Maybrat, Maranungku = Pitjantjatjara, Apurinã, Arapesh (Mountain) = Lepcha, Mandarin, Khmer, Georgian = Latvian, Rapanui, Khmu', Berta, Finnish = Persian, Awa Pit, Zuni, Martuthunira, Oromo (Harar), Spanish, Grebo = Malagasy, Semelai, Amharic, Nivkh, Mumuye, Carib, Epena Pedee, Burmese, Malakmalak, Campa (Axininca) = Nasioi, Wichí, Greenlandic (West) = Zande, Japanese, Ket = Wardaman, Kawaiisu, Lakhota, Albanian = Doyayo, Kiribati, Pirahã, Yimas, Koasati, Hixkaryana, Slave, Khoekhoe, Wambaya, Maricopa, Sango, Pomo (Southeastern), Murle, Passamaquoddy-Maliseet, Kayardild, Burushaski, Kiowa, Hausa, Yuchi, Ngiti, Drehu, Hamtai, Ijo (Kolokuma), Trumai, Ju|'hoan, Yagua, Paiwan, Krongo, Kurdish (Central), Armenian (Eastern), Cayuvava, Aymara, French, Oneida, Atayal, Yapese, Haida, Lezgian, Navajo, Arabic (Egyptian), Jakaltek, Zulu, Cree (Plains), German, Klingon, Hunzib, Cubeo, Nez Perce, Wichita, Kutenai, Iraqw, Coos (Hanis), Nunggubuyu, Tsimshian (Coast), Wintu, Wari', Paumarí, //Ani, Tlingit, Squamish, Quileute, Abkhaz.
Created and maintained by Ivan A Derzhanski.
Last modified: 23 November 2014.