A distinctive features approach to Lojban phonology


Chapter 3 of The Complete Lojban Language (and to a lesser extent Chapter 4) contains a great deal of information about Lojban phonology. This information is mostly given in the form of lists and tables. It includes the definitions of valid and invalid consonant clusters and diphthongs and the rules governing which gism forms are blocked by existing gismu. When I first read it I found it difficult to see the regularities that I was sure were behind the data. So, for my own use I began work on the following preliminary attempt to use distinctive features to convey those regularities.

I have chosen to employ a set of features borrowed from An Introduction to Phonological Analysis (Marlett, 2001). Marlett intended for this set of features to be sufficient to describe the sound system of any human language. I suppose that it's likely that different linguists would propose differents sets of features; but I believe that this set is broadly representative, and that most others would not be significantly different.

The features used are:

sonorant [son]
characterized by resonance in the oral cavity; sounds that are [-son] are sometimes called obstruents
consonantal [con]
characterized by "at least moderate constriction of the vocal tract"
continuant [cont]
characterized by a lack of complete blockage of the vocal tract
nasal [nas]
characterized by air flow through the nasal cavity
voice [voi]
characterized by vibration of the vocal folds
labial [lab]
characterized by involvement of the lower lip
coronal [cor]
characterized by involvement of the front (or crown) of the tongue
distributed [dist]
(dependent on coronal) characterized by involvement of the tongue blade as opposed to the tongue tip
lateral [lat]
(dependent on coronal) - characterized by a lowering of one or both sides of the midsection of the tongue
strident [strd]
(dependent on coronal) - characterized by more noise than other sounds
dorsal [dor]
characterized by involvement of the body (dorsum) of the tongue
high [hi]
(dependent on dorsal) - characterized by a raising of the tongue body
back [bac]
(dependent on dorsal) - characterized by a retraction of the tongue body
low [low]
characterized by a lowering of the tongue body

The following is a summary of the features of Lojban.

.sonconcontnasvoilabcordistlatstrddorhibaclowCLL description
`--+-----------an unvoiced glottal spirant
a+-+-+--------+an open vowel
b-+--++--------a voiced bilabial stop
c-++---++-+----an unvoiced coronal sibilant
d-+--+-+-------a voiced dental/alveolar stop
e+-+-+---------a front mid vowel
f-++--+--------an unvoiced labial fricative
g-+--+-----+++-a voiced velar stop
i+-+-+-----++--a front close vowel
j-++-+-++-+----a voiced coronal sibilant
k-+--------+++-an unvoiced velar stop
l+++-+-+-+-----a voiced lateral approximant (may be syllabic)
m++-+++--------a voiced bilabial nasal (may be syllabic)
n++-++---------a voiced dental or velar nasal (may be syllabic)
o+-+-+-----+-+-a back mid vowel
p-+---+--------an unvoiced bilabial stop
r+++-+-++------a rhotic sound [(may be syllabic)]
s-++---+--+----an unvoiced alveolar sibilant
t-+----+-------an unvoiced dental/alveolar stop
u+-+-+-----+++-a back close vowel
v-++-++--------a voiced labial fricative
x-++-------+++-an unvoiced velar fricative
y+-+-+-----+-++a central mid vowel
z-++-+-+--+----a voiced alveolar sibilant

Notice how the features [son] and [con] divide all Lojban sounds into four categories. All of the "regular" consonants are [-son +con]. All of the "syllabic" consonants are [+son +con]. The vowels are [+son - con]. The only sound that is [-son -con] is /'/.

It is not clear to me whether Marlett intended the features [back] and [high] as applied to consonants to be distinct from the features [back] and [high] as applied to vowels. I don't see any harm in treating them as the same features.


In the following discussion, C will represent any sound which is [+con] and V any sound which is [+son -con].

Chapter 3, Section 6 of The Complete Lojban Language defines Permissible Consonant Pairs (although that term is not actually used until Section 7). That definition can be restated as:

A consonant pair, C1C2, is a Permissible Consonant Pair (PCP) unless:
A1. C1 = C2
A2. C1 is [-son +x:voi] and C2 is [-son -x:voi]
A3. C1 and C2 are both [+strd]
A4. Cx is [+cont +dor] and Cy is [-voi -lab +x:cor +x:dist] (where either x = 1 and y = 2, or x = 2 and y = 1)
A5. C1 is [+son +lab] and C2 is [+voi -dist +strd]

Rule A1 says that doubled consonants are not allowed. Rule A2 is the "voicing harmony" rule. The notation here may need a little explanation. It indicates that whatever the value of the feature [voi] for the first consonant, the second consonant has the opposite value. Rule A3 says that the two consonants cannot both be from the set "c", "j", "s", "z". Rule A4 says C1C2 cannot be any of "cx", "xc", "kx", or "xk". Rule A5 says that C1C2 cannot be "mz". Note that Rules A4 and A5 are stated as one rule in The Complete Lojban Language. They appear to have little relation to each other.

Chapter 3, Section 7 lists forty-eight Permissible Initial Consonant Pairs. The content of that list can be explained by the following rules:

If a consonant pair, C1C2, is a PCP then it is also a Permissible Initial Consonant Pair (ICP) unless:

B1. C1 is [-cor] and C2 is not [+son +cont]
B2. C1 is [+son -lab]
B3. C1 is [-cont +cor] and C2 is [-cor]
B4. C1 is [-cont +cor] and C2 is [+lat]
B5. C1 is [+voi +strd] and C2 is [+son -lab]
B6. C2 is [+cont +dor]

Section 7 also gives rules regarding consonant triples. These rules only need slight modification:

A consonant triple, C1C2C3, can occur medially if:
C1. C1C2 is a PCP;
C2. C2C3 is an ICP; and
C3. it is not the case that C1 is [+nas -lab], C2 is [-cont +x:voi -cor], and C3 is [+x:voi +strd].

Chapter 3 Section 4 defines four classes of diphthongs. The first class of diphthongs are freely used in most Lojban words. The second class are used as stand-alone words (cmavo) and in cmene and fu'ivla. The third class are only used in cmene. The fourth class are not used in Lojban words.

D1. A pair of vowels, V1V2, is in the first class if either:

  • V1 is [+hi -low] and V2 is [+hi -bac], or
  • V1 is [-bac +low] and V2 is [+hi]

D2. A pair of vowels, V1V2, is in the second class if V1 is [+hi] and V2 is not [+bac +low]

D3. A pair of vowels, V1V2, is in the third class if V1 is [+hi] and V2 is [+bac +low]

D4. All other diphthongs are in the fourth class.

Chapter 4, Section 14 gives a table of which consonants were considered "too similar" to which other consonants for the purposes of the gismu creation algorithm. This information can be presented more systematically as:

E1. [+son +x:nas] blocks [+son +x:nas]
E2. [-cont +cor] blocks [-cont +cor]
E3. [+bac] blocks [+bac]
E4. [-son +x:cont -x:voi +lab] blocks [-son +y:cont +y:voi +lab] and vice versa
E5. [+x:voi -x:dist +strd] blocks [+y:voi +y:dist +strd] and vice versa


Marlett says (p. 224) "features have been used as part of a simplicity metric; that is, they have been used to formally measure the simplicity of a phonological rule. A natural process (one that is commonly found in languages) should be statable with fewer features than a process which is unnatural." My results seem to indicate that Lojban's phonology is, on the whole, satisfyingly "natural".

I welcome comments and suggestions for improvements.

Created by earendur. Last Modification: Thursday 19 of July, 2012 15:05:08 GMT by gleki.