Do Reviews in Other Scripts Score Sri Lankan Hotels Lower? (270,756 reviews)

Published: November 5, 2026 · 270,756 reviews analysed

Here is a pattern that shows up in almost every destination on this island: the reviews that give a Sri Lankan hotel its worst scores are, disproportionately, written in a script almost nobody at that hotel can read. Not in English. Not in Sinhala. In Cyrillic, Japanese, Korean, Mandarin, Arabic.

We put every one of the 270,756 reviews in our corpus that contains actual text into three buckets by script alone, then compared the scores. The English-script reviews average 8.7. The reviews written entirely in another script average 8.13. Reviews that mix English words into another script — a Japanese review naming an English hotel, a Russian review naming a staff member in Latin letters — average 8.15, the lowest of the three.

The obvious reading is wrong. We checked it.

The short answer

A review written in a script the property cannot read scores about 0.57 points lower than an English one — but the gap is not really about language, or about English-speaking guests being friendlier. It is about how much detail a review carries. About 32.4% of fully non-Latin reviews are three words or fewer ("very good", "نعم", "Отлично"), and 27.5% of the mixed-script ones, against 5.9% of English reviews, and the short ones drag the average down. Where a property has staff who can genuinely answer a Japanese or Russian guest — Mango House in Ella, Marino Beach in Colombo, Iwamisou in Sigiriya — the 0.57-point gap almost entirely vanishes. The clearest fix is not better service. It is one sentence of communication at check-in.

What we actually measured

We took all 363,148 reviews in the BYC.lk corpus and dropped two things: reviews with no text, and the Booking.com placeholder string "There are no comments available for this review", which appears thousands of times and is not a review. That left 270,756 usable reviews.

We did not try to detect language — language detection on a corpus this varied is unreliable, and a review in English with a Cyrillic surname would fool it. We detected script instead, which is exactly the practical problem: a front-desk team that reads English and Sinhala cannot read a Japanese review even if it is glowing.

The three buckets

Inside the non-Latin group the scripts split 1,880 Cyrillic (Russian, Ukrainian, Kazakh), 1,497 CJK (Japanese, Chinese, Korean) and 468 Arabic. All three sit below the English average, which argues against the gap being a quirk of one nationality.

Why this is probably not what it looks like

Three checks, and each one weakens the "unhappy foreign guests" story.

1. It is not one booking platform. The corpus mixes Booking.com and Agoda reviews. On Booking.com English reviews average 8.69 and non-Latin 8.07. On Agoda they average 8.81 and 8.72 — a gap of 0.09. Agoda's non-Latin sample is small (337 reviews) so treat it as a hint, not proof, but the pattern does not survive switching platforms intact.

2. The non-Latin reviews are far shorter. 32.4% of the fully non-Latin reviews and 27.5% of the mixed-script ones are three words or fewer, against 5.9% of English reviews. Short reviews score lower across this entire corpus — we found the same effect in our study of review length. A guest who writes "نعم" is not necessarily angrier than a guest who writes a paragraph; they are just leaving less.

3. The gap collapses where the staff speak the language. This is the part worth acting on. The properties with the most non-Latin reviews in the corpus show almost no gap where the house is built for that guest:

Mango House Japanese Guest House in Ella — a guesthouse with Japanese ownership — averages 9.77 across 129 non-Latin reviews and 9.80 across 271 English ones. Marino Beach Colombo, which has Japanese-language service, splits 9.14 and 9.19. Iwamisou in Sigiriya, the same. Where the property can actually respond to the review, the language penalty is roughly zero.

Compare that with Dutch Heritage Colombo: 47 non-Latin reviews averaging 7.87 against 8.67 for English, a 0.80-point gap. Or Heritance Kandalama, where the non-Latin group actually scores higher (8.86 against 8.49) — proof that this is not a law of the corpus, it is a property-level outcome.

Where the gap is worst

Across 25 destinations with enough reviews on both sides to compare fairly (at least 100 non-Latin and 300 English):

Matara has the widest gap of any destination on the island: -1.52 points, 7.27 against 8.79 across 101 non-Latin reviews. Polonnaruwa (7.85 vs 8.97) and Yala (7.7 vs 8.76) follow. These are the destinations where a non-English-speaking guest has the most to lose — the cultural sites, the safari camps, the towns where English is least common behind the counter.

Notice what is at the other end. Trincomalee shows a gap of just +0.25, and Hikkaduwa (-0.19) and Kandy (-0.21) are close behind. These are places with the deepest tourism infrastructure and the highest density of staff who have dealt with a Japanese tour group before.

What this means if you are booking

Ask whether anyone at the property speaks your language before you book, and ask in writing. Not because service is worse — because the review you leave, and the review you read, will both be shaped by whether the communication worked. If you write reviews in Japanese or Russian, the single most useful thing you can add is one line in English. It is the difference between a review the next guest can act on and a score with no explanation attached.

Do not read a 8.13-average non-Latin cluster as a red flag on its own. Check whether the English reviews for the same property agree. Where they diverge by close to a point, that divergence is the signal — it usually means one group was communicating and the other was not.

If you are a property owner, the fix is cheap. A printed one-page guide in Japanese, Russian and Arabic covering check-in time, breakfast hours, Wi-Fi, the water situation and how to call a tuk-tuk addresses exactly the complaints that appear in short non-Latin reviews. You do not need a translated website. You need the seven facts a guest needs in their first hour, in their script.

WATCH OUT FOR

This is a comparison of script groups, not a controlled experiment. The two groups are different people travelling differently: many non-Latin reviews come from group tours, and group-tour guests rate differently from independent travellers. We cannot separate the two, and we are not claiming to. What we can say is that the gap is real, it is large, and it disappears at properties with multilingual staff — which is consistent with communication being the mechanism, and inconsistent with it being about the guests themselves.

Small samples at the destination level. Matara's -1.52 rests on 101 reviews. Polonnaruwa, Yala and Tissamaharama are all over 100. Treat the ranking as directional and the direction of the whole table as the finding.

Short reviews are under-weighted here, not over-weighted. Because a three-word review still counts as one review, the short non-Latin reviews have exactly the same influence as a paragraph would. If we weighted by length the non-Latin averages would likely rise. We chose not to, because that would be us manufacturing a friendlier number.

GOOD TO KNOW

The island average is 8.7 for reviews with text. The non-Latin average of 8.13 is 0.57 points below — which is about the size of the difference between the island's best and worst coasts. It is not noise.

Mixed-script reviews score lowest of all three. 8.15, below fully non-Latin. These are usually reviews naming an English-named hotel or staff member inside another script, and they are also the longest non-Latin group (30.6% run over 50 words). The pattern fits a guest who engaged most and was disappointed most.

Trincomalee is the one destination that runs the other way, with non-Latin reviews scoring higher (8.53 vs 8.27). It is the only one of 25.

This study is part of a series. If you are choosing where to stay, start with what 314,000 reviews say about Sri Lanka, then read the most overrated and underrated destinations by review data. If you have already booked, what the check-in sentence tells you is the follow-up.

Booking a stay and unsure how something will be handled? Ask BYC at byc.lk — free, and honest about the trade-offs. Some links are affiliate links and never change our advice.

Data: BYC.lk analysis of all 270,756 guest reviews with usable text in a corpus of 363,148 verified hotel and guesthouse reviews of Sri Lankan properties (Booking.com and Agoda). Script classification by Unicode block; scores are on the platform 10-point scale as published. Methodology limitations are stated in the article and are not corrected away.