Manglish, Hinglish and the comments translate-first tools miss

Manglish feedback analysis means reading answers typed in Malayalam with English letters, often mixed with English mid-sentence, without translating them first. Most feedback from India isn't typed in one language. It's Malayalam or Hindi in English letters, switching to English for a word or two. Tools that translate first lose the half that was never English, so the complaint never counts.

How people actually type feedback

Ask a room in Kochi or Delhi what they thought, and the answers look like this:

• English: "The registration form timed out twice before it took."
• Manglish (Malayalam in English letters): "Registration form slow aanu, randu thavana try cheythu."
• Malayalam script: "Lunch-ന് vegetarian option എന്താ ഇല്ലാതിരുന്നത്?"
• Hinglish (Hindi in English letters): "Speakers ekdum zabardast the, but wifi पूरे दिन बंद रहा।"

All four are normal. Phone keyboards default to English letters, people switch language for technical words, and nobody stops to pick one. A feedback tool has to read all four.

Why so much feedback is typed in English letters

It isn't a lack of literacy. Most people in India learned to type on phones whose keyboards default to English letters, and switching to a Malayalam or Hindi keyboard mid-message is slow. So people write their own language phonetically: "venam" for "want", "aayirunnu" for "was", "bahut" for "very". There is no standard spelling, so the same word appears as "valare", "valre" and "velare" in one set of answers.

Then there's code-mixing: switching language inside a sentence. Technical and workplace words stay English ("registration", "slides", "deadline"), while the feeling and the grammar stay in the speaker's language. "Assignments deadline okke ore week-il aayirunnu" is a perfectly natural sentence to a Malayali student, and it carries one clear complaint: every deadline fell in the same week.

A feedback tool that only reads clean English, or only reads clean native script, misses the way your people actually write.

What translate-first does to "slow aanu, randu thavana try cheythu"

Many tools translate every answer into English and then analyze the English. For native script that mostly works. For Manglish it fails, because translation services treat Romanized Malayalam as broken English and return something like "Registration form slow aanu, two times try did." The meaning, "it was slow and I had to try twice", is gone, and the answer lands nowhere.

Why keyword tools see three different complaints

Keyword counts have the opposite problem. "Registration slow", "regstration form valare slow aayirunnu" and "रजिस्ट्रेशन फ़ॉर्म बार-बार अटक रहा था" are the same complaint, but they share almost no words. A word cloud shows three small mentions instead of one big problem, and the problem you most need to fix looks minor.

Worked example: one complaint, four ways of typing it, one theme

Sample answers written for this article; the analysis is the engine's unedited output.

In a sample of 30 answers from a partner summit, people complained about registration in four languages. The AqiloPulse engine read each in the language it came in and counted them in one theme:

As written	Language
"The registration form timed out twice before it took. I nearly gave up."	English
"Registration form slow aanu, randu thavana try cheythu."	Manglish
"रजिस्ट्रेशन फ़ॉर्म बार-बार अटक रहा था, बहुत समय लगा।"	Hindi
"نموذج التسجيل توقف مرتين قبل أن يعمل."	Arabic

Result: one Registration theme, 8 people, negative, the largest theme on the page. The breakdown by language sits beside it, so you can see which groups raised it.

The same happened with wifi: "Wifi kittiyilla, slides download cheyyan pattiyilla" ("no wifi, couldn't download the slides") was counted with the English complaint about wifi in the main hall.

Two subjects, two feelings, one sentence

Mixed-language answers often carry more than one opinion. "Speakers ekdum zabardast the, but wifi पूरे दिन बंद रहा" says the speakers were excellent and the wifi was down all day. Read as a whole it's "mixed", which is useless. Read properly, it's praise for the speakers and a complaint about the wifi, and each should count in its own theme with its own feeling.

That's how the AqiloPulse engine reads it: each subject in an answer gets its own feeling, so a speaker theme isn't dragged down by a wifi complaint that happened to share a sentence.

Reading the readout in Malayalam or Hindi

The readout itself can be translated for your team: the summary, the points and the theme names in Malayalam, Hindi and many more languages, with two rules that never bend. Counts stay counts, and quotes stay in the respondent's own words, untranslated.

Checklist: running a survey for a mixed-language audience

• Write the questions in the language most of your audience reads comfortably. One language per pulse is enough; people can answer in any language.
• Don't ask people to "answer in English". You'll get fewer answers and politer ones.
• Keep one open question per topic, so a mixed answer still has one subject.
• Choose the readout languages your team reads (Settings › Insight languages).
• When you share results, quote people as they wrote, with a translation beside it if needed.

Frequently asked questions

Can respondents answer in any script?

Yes. Each answer is read in whatever language and script it was typed, including Malayalam, Hindi, Tamil and Arabic in their own scripts, and Romanized Malayalam or Hindi.

Is Hinglish supported, or only Manglish?

Both. Romanized Hindi and Malayalam are part of how the AqiloPulse engine is tested, alongside native-script Hindi, Malayalam, Tamil and Arabic.

Do quotes get translated?

No. Quotes always appear exactly as written, labelled with their language. Only the summary wording is translated for your team.

Related reading: how to analyze open-ended survey responses and inside an AqiloPulse readout.