By now I am already 35 years in Cambodia, and my Khmer language level is still very limited. Of course I tried to find reasons to cover my laziness – especially my 92 years of age. I am of course aware that I am not the only one with this struggle. But to my surprise I found that also many Cambodian people avoid to use Khmer, that is: the WRITTEN Khmer language. The administration of Facebook – internationally one of the widest used communication programs – found out in 2018 that 50% of all Facebook Messenger’s worldwide Voice Mail came from one country: Cambodia. Nobody at the company could understand why.

Well, some thought because Cambodia is a small and poor country – it might have something to do with a generally low literacy level? But the Cambodian adult literacy level in 2018 was estimated to be 83%!

Facebook even made a special study with users of Messenger, Telegram, and WhatsApp in the countries of Benin, the Dominican Republic, Ivory Coast, and Senegal, asking why people used Voice Mail instead of written communication. 87% of the respondents said that they use voice tools only sometimes, to send notes in a different language from their own. A 2016 USAID report had shown that voice messages were popular because typing is difficult on the small screen of a mobile phone, more so when using the Khmer script.

But the fact that 50% voice mail world wide is from Cambodia seems to say something more about Cambodia. Typing is time consuming and difficult everywhere, but only in Cambodia a very high percentage people prefer to use the personal, easy way of sending and receiving voice messages.

Even without using scientific linguistics, some of the problems can be shown: In Khmer script, there are difficulties how phonetic language elements – phonems – relate to their graphic representation – graphems – in writing..

There are 33 Consonants forming the basis of Khmer syllables. Khmer consonants are sorted in two “Series”. Vowels change sound, based on the “Series” of the initial consonant. For example:

– the /k/ sound is written in the first Series as ក and combined with the written vowel ា as កា and pronounced as /kaa/

– the same /k/ sound is written in the second Series as គ (different, but only slightly) and combined with the same written vowel ា as គា, but pronounced as /kie/

In voice mail the sound for the /k/ consonant is the same in both Series – it is combined with the same vowel ា if it were written, but the vowel sound we hear is either /aa/ or /ie/. – No need to consider this in voice mail, but only when writing.

Khmer characters map to phonemes, with variations based on position (initial, medial, final) and Series. Below are key categories and examples.

Consonants can be part of Clusters in the following examples: Two-consonant Clusters at the start of words (ក្ន), or Clusters within words (ក្ល), or word-final Clusters (កសិកម្ម). But there are also Three-consonant Clusters (ង្ក្រ).

This system of combining written consonants is complex, because written Khmer uses consonant combinations, putting some combined consonants above, below, before, and after the base character, so that such a “character” can visually span three vertical zones. And the visible sequence of written letters may be different from the sequence of the spoken sounds. Lets see an example: ស្រី /srəj/ “female”. Where are the written consonants, and where is the vowel paced?

ស្រី

əj

r s

And how are the letters read read: s–r–əj

Some such complex “Characters” may have pre-positioned consonants (ស្រី /srəj/ written as r-s-əj). Or there may be a fully pre-positioned vowel (គេ /ke/ “they” written as “ek”), or even part of a split pre-positioned vowel graphem written before a consonant (ទៅ /t aw/ “to go”), and the different graphem of the other part of the same split vowel is written after the consonant.

What about the written Vowel system in Khmer? There are Dependent (23) and Independent (5) Vowels.

Dependent Vowels cannot stand alone, in writing they are attached above, below, left, or right of the related consonant. Their sound depends on the series of the related consonant series:

Example 1: ា /a/ after Series A, but /iə/ after Series O.

Example 2: ុ /o/ after Series A, but /u/ after Series O.

For a description of the Khmer vowel system I change here the font size, in order to better show details of the Khmer writing.

Dependent Khmer Vowels, with phonetic IPA (International Phonetic Alphabet) values, and positions:

Above Consonant

• ិ → /i/ (short)
• ី → /iː/ (long)
• ឹ → /ɨ/ (short, central vowel)
• ឺ → /ɨː/ (long)

Below Consonant

• ុ → /u/ (short)
• ូ → /uː/ (long)
• ួ → /ua/

Before Consonant

• េ → /eː/
• ែ → /ɛː/

After Consonant

• ា → /aː/
• ោ → /oː/
• ៅ → /aw/

Surrounding Forms

• ៀ → /iə/
• ឿ → /ɨə/
• ៀ (variant) → /ie/

To remember the order Consonant → Vowel → Diacritic is important in handwriting and in OCR analysis.

Independent Vowels stand alone as full syllables.

Examples: ឥ (/i/), ឦ (/ii/), ឧ (/u/), ឯ (/e/), ឰ (/ai/).

As the same graphic vowel sign changes the sound, depending on the Series of the related consonant, relations have to be learned for correct writing.

In addition to consonant and vowel letters, there are Diacritics in written Khmer. Diacritics (វណ្ណយុត្តិ) serve essential functions to modify vowels, or alter consonant Series, or act as orthographic or tonal markers. Khmer consonants belong to two Series: 1st Series (inherent -â /ɑː/ sound) and 2nd Series (inherent -ô /ɔː/ sound). Just two examples among the eight that exist:

OK, is it by now clear why so may people in Cambodia prefer to send voice messages, and not write their messages?

But relying on the easiness of voice messages creates other problems for communication.

What was written, can be easily revisited for clarification, or be taken up for a new round of communication – not so in voice mail. Voice conversations are short-lived. One cannot scroll back to recall details of an exchange some minutes, or some days ago, one can only refer back by remembering what had been said in a voice messages. It is impossible to use search functions to get deeper involved by discussing later what had been mentioned in a former chat.

Therefore political statements and business and other messages, where detailed facts and numbers are important, in Cambodia tend often to be written in specific subject jargon, or often in English.

It is not only the general convenience of speaking instead of writing, that leads to the internationally exceptionally high use of voice mail in Cambodia – also the special complications of the Khmer writing system make it exceptionally difficult to develop computer based Machine Translation programs for the Khmer script.

It is not only difficult to WRITE using the Khmer script, but is is also extremely difficult for Optical Character Recognition (OCR) computer programs to READ Khmer writing – an important initial step for Machine Translation applications.

Some of the challenges:

– Khmer writing does not use spaces between words, but only at the end of sentences.

– OCR does not see Words, but long flows of “Characters” – but a “character” may visually span over three vertical zones (above, normal, below), including pre-positioned or past-positioned parts..

– OCR programs must recognize these as units, not as separate letters. 

– OCR programs must learn word units, without being delimited by free spaces, from linguistic patterns.

– Khmer characters have non-uniform widths, curved shapes, and visually similar glyphs.  Many letters differ only subtly (example: ក and គ), increasing misreadings.  

– There are not many Khmer written resource materials compared to the large amount of written material in many other languages; Khmer lacks large, diverse, high-quality datasets used to create LargeLanguageModels needed to develop Artificial Intelligence systems, necessary to develop Machine Translation programs. (Such have been developed for other languages, differently for printed or hand written text and street signs etc.) 

– Diacritics and ligatures in Khmer script can appear in multiple positions around a base character.  OCR must detect tiny marks that may be faint or touching other written glyphs.

– And there are other problems, related to the Khmer language itself: in Khmer often pronouns (“I/you/he/she/we/they”) are not always stated or repeated, and Machine Translation may misinterpret the context; idioms, and cultural references may not be understood and often result in nonsensical texts.

While professional scientific research to develop Machine Translation programs continues, actual translation work is using existing OCR and Machine Translation programs with special restriction:

– To be translated Khmer text has to be cut into small sections, maybe not much longer than one printed book page.

– Make two translations with two different Machine Translation programs, and compare both results with each other, and the original text, doing a human verification and necessary corrections.

– Repeat the same process for all sections, using any insights from one section, when doing the translation and the human verification for the next, where necessary revising also previous sections, until the whole text is completed, and a total human review of the complete text is achieved.

The fact that the production of Khmer texts is so difficult has tremendous results:

– Compared to the production of written literature in countries like Japan or Korea, there are not many books being written in Khmer, which could result in more intellectual communication among the Cambodian public.

– Where there are articles and books written for their own public, but translated into other languages – then the international community can know a lot about Japan and Korea.

– As far as I know – but I may be wrong – there are no Khmer books, written in modern time, by Cambodian authors written for the Cambodian public, and then translated into other languages. What is known internationally about Cambodia is based mostly on books and substantial articles written by foreigners in foreign languages, or on publications written by Cambodians in foreign languages.

There is an example, where an older Khmer text attracted international attention – about Cambodian social rules and values. But this is not an example where Cambodians wanted to share their values internationally – it is an example how the Chbab Srey (“Female Law”) was taken up by foreign interests, so that it led to limited translations, by foreigners.

The Chbab Srey is attributed to King Ang Dong around 1837. It had been transmitted orally for centuries, before being written down formally. There were some French translations, English had only partial academic translations, interested in traditional Cambodian gender roles, emphasizing female obedience and domestic responsibilities – as gender studies and historical analysis.

It has contributed to the “image” of Cambodia; the changed role, by now, of this document in Cambodian education, and the present gender related discussions in Cambodia are internationally not much known,

The international community has no access to the public communication about values in society, about what is going on in Cambodia in the Khmer language. Because of the limited possibility of Khmer Machine Translations, participants in the public discussion, in the Khmer language, can not hope that their interests, concerns, vales, and arguments, will be noted abroad.

This situation, based on the Khmer language, resulting in limited international serious communication on values and motivations, is not discussed publicly, and it is not easy to overcome it by using Machine Translation.

What possibilities exist or could be developed, to overcome this intellectual international isolation of Cambodia?

Any responses towards a public discussion are welcome.

Norbert Klein – nhklein@gmail.com