Jump to the Bug Reports/Observations with the new translation engine section
Introduction
A custom machine translation tool for the Latin American and Caribbean Regional At-Large Organisation (LACRALO) mailing lists was implemented around early 2011. However several issues or factors continue to negatively impact the working of the machine translation tool and in turn, has lead to great difficulty in communication and collaboration with the English and Spanish speaking communities in the LAC region.
See presentation by the At-Large Technology Taskforce Working Group for the ICANN53 meeting which has worked to identify the issues and continues to follow this issue with ICANN Staff.
To recap, LACRALO has two mailing lists
- LACRALO list in English: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/
- LACRALO list in Spanish: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/
Emails in english sent to lac-discuss-en@atlarge-lists.icann.org are machine translated via your custom tool using Google Translate and posted to lac-discuss-es@atlarge-lists.icann.org.
Similarly, emails in Spanish sent to the lac-discuss-es@atlarge-lists.icann.org are translated and posted to lac-discuss-en@atlarge-lists.icann.org.
Issues identified since 2012
(this section is from a June 18 2012 email)
1) Attachments in emails sent to a list are not received on the other list.
When an email with attachments such as PDFs is sent to one list, the subject line and body of the email is translated and sent to the other list BUT without the attachment.
2) Subject lines of translated emails from ES to EN become garbled.
The subject line of translated emails (seemingly) from the lac-discuss-ES list to the lac-discuss-EN list often translated to garbled text.
Examples abound from a review of the archives.
(a) First email posted to lac-discuss-en list :
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005932.html
Subject line: [lac-discuss-en] ICANN full list of applied for gTLD strings
(b) which is translated and posted to lac-discuss-es list as:
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004552.html
Subject line: Lista completa de la ICANN solicitó cadenas de gTLD
(c) Someone on the lac-discuss-es list responds posts to lac-discuss-es list as:
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004553.html
Subject line: [lac-discuss-es] Lista completa de la ICANN solicitó cadenas de gTLD
(d) which is translated and posted to the en list as:
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005933.html
Subject: [lac-discuss-en] =? Iso-8859-1? Q? Lista_completa_de_la_ICANN_solici? == Iso-8859-1? Q? T = F3_cadenas_de_gTLD? =
Another example:
Email on lac-discuss-es list : http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004518.html
Subject line: [lac-discuss-es] RES: Alerta de Noticias de la ICANN - Aviso de Prórroga del período que abarca la ICANN: ICANN FY13 Proyecto de Plan Operativo y Presupuesto
gets translated and posted as an email on lac-discuss-es list: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005897.html
Subject line: [lac-discuss-en] =? Utf-8? Q? RES = 3A_Alerta_de_Noticias_de_la_ICANN_? == Utf-8? Q?-_Aviso_de_Pr = C3 = C3 = ADodo_que_abarca_la_ B3rroga_del_per =? == Utf-8? Q? ICANN 3A_ICANN_FY13_Proyecto_de_Plan_Operativo_y_Presupu =? == utf-8? q? this? =
Note the difference with "Utf-8? Q?" in this example as compared to "Iso-8859-1? Q?" in the previous example.
Such gibberish in the subject lines can get even worse if someone responds on the EN list and the translation further scrambles the subject line on the other list.
Again, examples abound from a review of the archives but as one example, consider the subject line for an email on lac-discuss-es list
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004039.html
Subject line: [lac-discuss-es] =? Iso-8859-1? Q? Invitación = F3n_a_la_reuni = F3n_/_LAC? == Iso-8859-1? Q? RALO_Costa_Rica_Eventos_rueda_de_prensa_Grupo_de_Tr? == Iso-8859-1? Q? Abajo_el_martes_06_de_marzo_2012_a_las_20 = 3A00_UTC? =
which gets translated and posted to the EN list as
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005357.html
Subject line: [lac-discuss-en] =? Iso-8859-1? Q? = 3D = 3F_Iso-8859-1 = 3F_Q = 3F_Invitac? == Iso-8859-1? Q? I = F3n_ = 3D_F3n = 5FA = 5Fla = 5Freuni_ = 3D_F3n = 5F / = 5FLAC = 3F? == iso-8859-1? q? _ = 3D = 3D_Iso-8859-1 = 3F_Q = 3F_RALO = 5FCosta = 5FRica = 5FEv? == iso-8859-1? q ? ents = 5Frueda = 5Fde = 5Fprensa = 5FGrupo = 5Fde = 5FTr = 3F_? == iso-8859-1? q? = 3D = 3D_Iso-8859-1 = 3F_Q = 3F_Abajo = 5Fel 5Fmartes = 5F06 =? == iso-8859-1? q? = 5Fde = 5Fmarzo = 5F2012 = 5FA = 5Flas = 5F20_ = 3D_3A00 = 5FUTC? == iso-8859-1? q? = 3F_ = 3D? =
3) Missing [lac-discuss-es] in subject lines of translated emails posted to the lac-discuss-es list
Consider example #1 again -
First email posted to lac-discuss-en list :
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005932.html
Subject line: [lac-discuss-en] ICANN full list of applied for gTLD strings
which is translated and posted to lac-discuss-es list as:
Email: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004552.html
Subject line: Lista completa de la ICANN solicitó cadenas de gTLD
The email at http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004552.html shows that the subject line is missing the [lac-discuss-es]. This hampers filtering by ES users and makes it difficult to track threaded conversations.
4) Unusual superscript and other odd characters in translated emails
There have been numerous complaints about the quality of the translation of the actual body of emails with strange characters, some of which are superscript characters appearing in the translated version.
Examples aboud (repeating phrase, I'm afraid) on the LACRALO list archives, here is one example:
Email to lac-discuss-en : http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2012/005858.html
got translated to this on the lac-discuss-es list: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/2012/004483.html
As you can see,
* a character like a double quote " is translated to "
* a word like "organisation" is translated to organización
* a sentence like "The highest decision making body in any organisation is also subject to rules." is translated to
"El más alto órgano de decisión en cualquier la organización también está sujeto a reglas."
History of machine translation of LACRALO mailing lists
- The LACRALO list in English: http://atlarge-lists.icann.org/pipermail/lac-discuss-en/ and LACRALO list in Spanish: http://atlarge-lists.icann.org/pipermail/lac-discuss-es/ were machine translated around September 2008. The translation engine was based on Systran.
- The quality of translation left a lot to be desired in terms of the language translation. Also, there were technical issues such as the subject lines of translated emails becoming garbled.
- During the LACRALO meeting in Cartagena in 2010, there were intense discussions regarding the quality of mailing list translations. Some wanted to disable the translation entirely. Others wanted to keep the imperfect translation for mobile users and disabling translation would send the wrong signal to ICANN when ICANN should be giving language services a higher priority. ICANN staff acknowledged the issues and stressed that IT staff were aware of the issues and would be looking to solve the translation problems.
- Around May 2011, the translation engine was switched from Systran to Google Translate.
- in May 2011, Google announced that it would be depreciating the (then) free Google Translate API (http://googlecode.blogspot.com/2011/05/spring-cleaning-for-some-of-our-apis.html) in a few months but later (due to public outcry) announced that a paid version of the translate API would be offered.
- Other issues noted around June 2011 were identified, one was the translation engine would stop translating emails when they became too big. This meant that a conversation thread would stop being translated leading to missing emails from one list not being translated to the other list. See http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2011/004308.html for the solution to this (translation tool would stop translating quoted text)
- Around August 2011, the translation engine was switched back to Systran due to Google rejecting calls to the Google Translate API. See http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2011/004368.html
- Update to the translation engine: September 2011, http://atlarge-lists.icann.org/pipermail/lac-discuss-en/2011/004498.html
- June 2012, detailed analysis of translation issues (as noted in the introduction above)
Testing of new translation tool for LACRALO mailing lists
ICANN Staff have created two mailing lists (New-transbot-en and New-transbot-es) with a select number of persons on those lists for testing purposes.
Some of the key changes implemented in the new translation tool.
- The lack of punctuation was identified as a key issue for the poor translation of emails. This is because the translate tool can only send a certain amount of characters to the Google Translate API. Without punctuation, the translation tool would have to send text mid sentence.
One of the outcomes from the LACRALO translation WG was the Proposed Notice when email is not translated message which would be sent to the user if the email had formatting issues. - Subject lines would not be translated to ensure the conversation thread would not be lost and reduce the chance of garbled subject lines.
Testers of new translation tool for LACRALO mailing lists
| Name | Affiliation |
|---|---|
| Dev Anand Teelucksingh | LACRALO |
| Harold Arcos | LACRALO |
| Renata Aquino Ribeiro | LACRALO |
| Raitme Citterio | LACRALO |
| Humberto Carrasco | LACRALO |
| Seun Ojedeji | AFRALO |
| Satish Babu | APRALO |
| Glenn McKnight | NARALO |
| Silvia Vivanco | ICANN At-Large Staff |
Bug Reports/Observations with the new translation engine
Below are some of the bug reports/observations noted during testing
| Status | Date Added | Description | Additional Notes | |
|---|---|---|---|---|
| (Noted by Dev Anand Teelucksingh and Satish Babu) | 28 Feb 2016 | Subject line in body of translated email has the sender and first line of the email on the same line, when the sender and first line should be on separate lines. Two examples:
| The converted body text was included immediately after the subject and from line in translated emails. A new line character was inserted between sender name and first body text line to separate them. Initial testing complete; additional testing in progress. Example: https://community.icann.org/x/AofDAw | |
| (Noted by Satish Babu) | 28 Feb 2016 | There is an empty space in the beginning of most (but not all) lines in the translated email | Tested with mail IDs Outlook, Yahoo, Gmail. Verified that translated emails are not indented; empty spaces are not appearing at beginning of the lines Example: https://community.icann.org/x/vYPDAw | |
| (Noted by Satish Babu) | 28 February 2016 | At the end of the message, the sender's name starts with a lower case ('dev Anand', although it is 'Dev Anand' in the original message | Google Translate understands 'Dev' as an abbreviation and as a rule converts it to all lowercase Name was hardcoded to fix 'Dev' to begin with capital letter Tested with mail IDs Outlook, Yahoo, Gmail. Verified that name is appearing correctly with first letter capitalized Example: https://community.icann.org/x/ooPDAw | |
| (Noted by Satish Babu) | 28 February 2016 | 'Transbot' is mis-spelled as 'tansbot' (third line from the bottom) | Hardcode fix applied to correct spelling of 'tansbot' in translated emails Tested with mail IDs Outlook, Yahoo, Gmail. Verified that spelling is now appearing correctly as 'transbot' Example: https://community.icann.org/x/oILDAw | |
| (Noted by Dev Anand Teelucksingh | CRITICAL | April 17 2016 | the transbot can't handle cedilla - as At-Large Staff signature lists a staff member with a cedilla in her name, any message from At-Large Staff will result in a message not translated
The April thread with the subject line "CALL FOR MEMBERS: At-Large Public Interest Working Group" on EN : http://mm.icann.org/pipermail/new-transbot-en/2016-April/thread.html and ES : http://mm.icann.org/pipermail/new-transbot-es/2016-April/thread.html showed how the issue was isolated after several variations of the original email were tried. This is a critical bug, as any cedilla in any word in an email would result in the email not being translated. | |
| Noted by Dev Anand Teelucksingh | April 17 2016 | The phrase "This Working Group is open to interested members of the At-Large community." gets translated to "Este grupo de trabajo está abierto a los miembros interesados \u200b\u200bde la comunidad de alcance." | The issue was related to zero-width space, which was being injected by Google Translate API. Zero-width space is used after characters that aren't followed by a visible space, but after which there may be a line break (source). It was encoded into the Unicode and was appearing as /u200b. To fix, it was replaced in the translated text with no space. Tested with mail IDs Outlook, Yahoo, Gmail. Verified that phrase is translated correctly without additional characters Example: https://community.icann.org/x/t4PDAw |
