Lojban.org claims that “Lojban has [...] unambiguous resolution of sounds into words”¹ and that “Lojban [...] sounds can be divided into words in only one way.”² What is the evidence for this? Is there a list of rules, algorithm, or program that one could use to segment any Lojban sentence with no spaces into words? Can this be done based purely on Lojban phonotactics, or does it require some knowledge of Lojban vocabulary? Does this require the sentence to be grammatically correct, or can it be a random mishmash of Lojban words?
This was an important consideration in the design of Lojban, so while I haven't gone through the process of verifying it myself, I'm fairly confident that it is true for most of the language. (Since just one mistake would make it false overall, it's hard to be certain that it's true for all parts of the language: there are a few difficult parts.) It's not extremely difficult to design a system like this for a constructed language, and Lojban has been around long enough that I imagine someone would have spotted any obvious errors in the original design of the phonotactics.
Note that in practice, the rules are only fully unambiguous as a mapping from a phonological/phonemic transcription to the sequence of words, not as a mapping from physical sounds (phones) to the sequence of words. Spoken Lojban, like any spoken language, is not immune to having auditory signals obscured by noise, or to having a certain phone being perceived as one phoneme by one speaker but as another phoneme by another (e.g. an English speaker listening to a Spanish speaker of Lojban might mishear as /b/ an unaspirated [p] phone that was intended as /p/).
I don't know exactly where the rules are laid out. My understanding is that no part of the rules requires knowledge of Lojban vocabulary: the rules work from the information included in the normal Latin spelling of Lojban, including the characters
. (which represent contrastive elements of Lojban pronunciation), plus stress. I'm not sure about whether there are any ungrammatical sequences of valid words that could break the rules, but I don't think so.
In Lojban texts written with spaces, stress is not typically written, but stress has to be provided for proper word division of unspaced spelling.
The contrastive element
. (the period), which is necessary for unambiguous parsing, is supposed to be pronounced as a "pause" (it looks like a lot of people interpret this as a glottal stop).
Proper nouns and borrowed words are subject to special restrictions for the purpose of avoiding ambiguity
There are some "hacks" (my characterization) to get around certain possible ambiguities.
Proper names, which might otherwise introduce ambiguous sequences of sounds, are required to end with a consonant followed by a
. (the character representing a "pause"), and
cannot used to not be allowed to contain the sequences
doi. In 2015, the la/lai/doi rule was officially abandoned in favor of a simpler rule stating that all names have to start as well as end with a
. "pause" (the new rule is called "dotside").
There are also complicated rules about what forms borrowed words can take.
Some rules about Lojban phonotactics can either be implemented multiple ways, or have changed over time. Here is a link to a discussion from 2015 with some more details:
And another link with discussion of some difficult points:
There is a potential area of ambiguity in Lojban phonetics with regard to something called a "buffer vowel".
A "buffer vowel" is an optional epenthetic vowel used to break up consonant clusters. It is intended for use by speakers of languages that have phonotactic limitations on consonant clusters (emphasis mine):
Many languages do not have consonant clusters at all, and even those languages that do have them often allow only a subset of the full Lojban set. As a result, the Lojban design allows the use of a buffer sound between consonant combinations which a speaker finds unpronounceable. This sound may be any non-Lojbanic vowel which is clearly separable by the listener from the Lojban vowels. Some possibilities are IPA [ɪ], [ɨ], [ʊ], or even [ʏ], but there probably is no universally acceptable buffer sound. When using a consonant buffer, the sound should be made as short as possible.
The buffer vowels appears to tend to be a high mid vowel, since that's an open space in the Lojban vowel system. But this appears to be a place where some ambiguity can arise:
- [ɪ] (buffer) is close to [i] (Lojban 'i')
- [ʏ] (buffer) is close to [y] (acceptable variant of Lojban 'i')*
- [ʊ] (buffer) is close to [u] (Lojban 'u')
- [ɨ] (buffer) is close to [ə] (Lojban 'y')
As "The Lojban Reference Grammar" puts it (emphasis again mine):
Since buffering is done for the benefit of the speaker in order to aid pronounceability, there is no guarantee that the listener will not mistake a buffer vowel for one of the six regular Lojban vowels.
The standard guidance seems to be to pronounce a buffer vowel short, and regular vowels long. I'm not a Lojban speaker, so I don't know well how this works in practice. Since the grammar does call out the potential ambiguity, I thought it made sense to mention in response to this question.
* roundness does not matter in pronunciation of Lojban vowels.