Ideas for improving local/international/english search modalities

CWagner

I gave an example: https://kagifeedback.org/d/3022-ideas-for-improving-localinternationalenglish-search-modalities/6

carl

Let me frame the question like this: Why do English results get pushed to the top on "International" when there is nothing indicating that the user even understands English? In my setup there is nothing in English, OS language, browser language, Kagi language, keyboard map. Why English then? Why not Mandarin or Russian?

Seeing English as a default is not the right perspective.

Vlad

carl English is de-facto language for international communications.

carl

Vlad Okay. I just wish it could be reverted to how it was, because I don't understand the need for two English filters. In Google I search for a term in English, I get English results, I search in Swedish I get Swedish results, I search in Finnish I get Finnish results, I search in Portuguese I get Portuguese results. Never having to manually set or change any region. Kagi "International" worked like this as well until recently, even better than Google.

Vlad

CWagner Would it be acceptable for you that all results are in german when you specify "de" region? (no english results). Because we can either force a mix results (and you have to be ready that an english result could ranked higher than a german one) or force language specific results (eg: only in german; but then when searching for kagi on "de" region, you would not get kagi.com)

kristofferR

Vlad Give options! Forcing local language only results can be great for occasional searches, or for those who prefer it permanently enabled like @CWagner, but it's a horrible default - especially for non-huge languages.

CWagner

Vlad Would it be acceptable for you that all results are in german when you specify "de" region? (no english results).

The way it used to be until recently was better, but only getting German results when searching with !de works for me.

Vlad We did not change how International works, so if there is any perceived 'change' we need to better understand what that is.

Until about last year, when there was a language mix, it was far more weighted in favor of the language you either selected by going regional, or by searching unambiguously in that language. What specifically brought that change. Maybe the people here who do regional searches more frequently than me can try and pinpoint the change.

Holger

Vlad we can either force a mix results (and you have to be ready that an english result could ranked higher than a german one)

I think people are used to mixed results and to English content possibly being ranked higher, so the discussion is 'just' about the extent. I guess this makes the issue so complex to discuss and analyze :-(

Until about last year, when there was a language mix, it was far more weighted in favor of the language you either selected by going regional, or by searching unambiguously in that language.

Kagi's results lately do seem a bit off to me as well, in this regard (compared to other engines). I mentioned elsewhere that I now see the case where for a local web site that translates some content to English, those translations are ranked higher than the local content. Similarly, en.wikipedia.org results are often ranked higher than their local equivalents. Those might be two examples of why the ranking feels wrong-ish from a user perspective?

What specifically brought that change.

The cause is obviously hard to tell from the outside. As for when things changed, my guess would be around December. I made this wild speculation in the posting linked above:

The results for those queries don't appear to be stable for me, in some cases the German-language results are preferred as I'd expect. I think I've read somewhere that you're currently using Brave for maybe 10% of the queries, but I can obviously only guess whether that might be related.

May well be nonsense of course (and if it happened to be true, then the Brave-influenced results would seem to work better for me). But either way I think the region-related ranking change happened somewhere around the time the Brave thing was announced.

Rasmus

Vlad I've read the thread, and would like to add my specific feature request, and the reasoning:

FR: Allow `lang:[xx]` to specify result language in search query

E.g.: commander cadre lang:fr

When I search in specific languages, it's not necessarily from the region where most people speak a certain language. It's to find documents whose existance I'm pretty sure of.

My search activity is 35 % legal, 35 % developer and 30 % personal.

A query example could be a position paper by Greenpeace, on the legality of occupying drilling rigs, written in Dutch. It may be on Greenpeace's Canadian site, for all I know. Most HTML pages have a lang tag, so it should be fairly easy.
Another example could be an article written in English, by a Swedish academic, teaching in Karlsruhe.
A third example is faux amis. commander cadre means very different things in French and English.

Vlad

carl We did not change how International works, so if there is any perceived 'change' we need to better understand what that is. The only thing we changed was inclusion of additional locales like 'CA - EN' and 'CA - FR' for a few regions.

zarwiz

Holger To me, such ideas seem unrelated to the question how the ranking should look like for the 'common' localized case.

I have the feeling this discussion mixes two different issues in one. Some people are concerned about language in regions where multiple languages are used by the same person and others how the ranking looks in ‘’common’’ cases.

Vlad The only thing we changed was inclusion of additional locales like 'CA - EN' and 'CA - FR' for a few regions.

How was Kagi interpreting results in Canada before this change? Was it english only? Was it multilingual? Did it behave more like international today? Is it just a change in regards to the geography?

Vlad

CWagner

but only getting German results when searching with !de works for me.

Note that this means if you searched for kagi or openai you would not get the (english based) homepages?

carl

How about a "Multilingual" option, where the user can just click checkboxes in the list for the languages she'd like included? If this would cost more on the backend, you could make it a paid Kagi add-on. I'd happily pay extra.

Holger

carl How about a "Multilingual" option, where the user can just click checkboxes in the list for the languages she'd like included?

To me, such ideas seem unrelated to the question how the ranking should look like for the 'common' localized case.

Vlad

zarwiz We were just using CA region which could mean anything. By being more specific (en/fr) the results are now more fine tuned.

zarwiz

Before, when it was only CA region, I don't remember Kagi making much distinction between english and french websites in Canada, maybe more english biased but overall pretty acurate given the language used in the querries, which made the experience much more organic. French canadians are used to this reality, and we consume english canadian website as well when browsing. I don't see how this is an improvement as french is a minority language in Canada, and completely ignoring french results when tuning in to EN-CA must not change much from a canadian perspective.

Maybe some people enjoy this new locale feature, but its made my experience horrible. Just today I had to manually change the locale more than 20 times. I see myself constantly using the g bang to get what I want quicker, while before Kagi felt much more relevant. I really hope this is solving other people's issues in Canada in some ways because the upside in Quebec is definitely hard to find.

MattLisson

Vlad

I agree that there are two conflicting schools of thought, but I would say that the results based on language and results based on region shouldn't necessarily be tied together.

As you say, international communication tends to default to english, but someone's language isn't necessarily tied to their region in many cases, so the best possible solution would be to ensure the backend treats language and region separately and then allows choosing different settings in the UI (Although probably too big an effort right now).

For my original query about shopping queries, its really a special case where I flip back and forth between the two schools of thought you post. I search for reviews world-wide for products that are international (Tech, Websites, etc), but only locally for local services and products (vegemite, mobile phone plans, etc). I'm not suggesting biasing the search engine one way or the other, but letting us choose.

Three suggestions for how we could choose this in the UI:

In the existing region drop down box, add an "Exclusively" button to the far right side, to allow selecting a region and excluding all other region's results.
- This requires the backend knowing that .com.au sites are local to Australia, .co.uk to UK, and so on.
- There are other cases where .com sites are hosted in Australia and have Australian local content, but skipping those is fine by me to keep it simple.
Using the multi-region UI mocked up by Teriy-Aki, display "International" as enabled by default (since international results are included by default), then selecting "International" will remove it from the selected regions and perform the same as option 1.
A separate toggle (possibly hidden in an overflow menu) that just says "Exclude international results".
- Again, toggling this on would perform as option 1.

In an ideal world, these same selections would be available separately based on languages, and the default settings for a search would be inferred from the language of the search terms. But I'll leave others to pitch that change.

benoit

Currently, the search results are more or less directly tied to the language of your search terms. If I search in English, I'll get English results. If I search in French, I'll get French results. English results are ranked against English results, French results against French results.

It would be good to have the ability to do multilingual search.

We could have an option in the settings to select languages we know how to read (and therefore would be happy to receive results from).

If I search, for example "Best Headphones", it would both search in English as well as the translated search terms "Meilleurs Écouteurs" for my selected languages and then display the results in all selected languages.

As an example, in English, I have "Head-Fi.org" (which is a headphone review site) and in French "Que Choisir" (which takes a 'scientific' approach to reviewing products), both raised. In a multilingual search, I'd then hopefully have both results show up instead of just the English raised result.
This way too, the Assistant could hopefully use the other language source to have even better quality inputs.

benoit

As a quick note, 54% of Europeans speaks at least two languages for example.
In countries like Luxembourg, Germany, Netherlands, or Belgium, it's generally three.

pimterry

I too have issues with this. Imo, the main issue is that language & location tied together. They are often not directly related.

In my case, I am British, but I live in Spain - a country with many international immigrants, multiple official local languages, and a primary language that's very widely used elsewhere.

I generally keep my Kagi configured for Spain (ES - es). Specific examples of where this causes problems:

Searching for globally relevant topics in Spanish returns results in Spanish, but worldwide - for example many US government websites are translated in Spanish, and I frequently find that searches for official information in Spanish returns the US gov's spanish-language info, Latin American discussions, or US spanish-language news. When searching in Spain, I want to prioritize Spanish regional info specifically, not Spanish language from everywhere. For example, searching for "impuesto ingreso" (literally "income tax") in "Spain ES-es" Kagi returns irs.gov/es/credits-deductions/individuals/earned-income-tax-credit/earned-income-and-earned-income-tax-credit-eitc-tables (the US tax agency) #1 and agenciatributaria.es/AEAT.educacion/Profesores_VT3_es_ES.html (the Spanish tax agency) #2. Adding "españa" as a search term fixes the issue, but this is annoying and redundant with the region is set.
Searching with Spain configured as the region (but my browser language in English, so the Kagi UI is all English) constantly returns only Spanish versions of international multilingual websites, such as Wikipedia or Mozilla Developer Network. If I search for 'CSS' then developer.mozilla.org/es/docs/Web/CSS (note the /es/) is the only MDN result. It should prefer the language I'm actually using - the region does not define the language.
Sites in co-official Spanish languages seem to be omitted entirely in some cases. For example searching for 'beteve' (Barcelona's TV channel & news network) with "Spain ES-es" region doesn't return their website (beteve.cat) on the two pages at all. Searching internationally it comes up as #2. I suspect this is because it's in Catalan - regionally Spanish but not linguistically.

There should be separate way to specifying your preferred language(s) and region. Note that Google and Bing do offer both as separate settings. Region detection for content is tricky, and could take in all sorts of complex signals, but should probably start by prioritising all related TLDs (.es, .cat, .barcelona) over others. Language preference should probably just come from the browser/OS imo which already provides this in depth (allowing for 1+ options and preference ordering, which is great) but it could also be a separate control too.

This is easily my biggest problem with Kagi and causes me issues every single day. I would really like to be able to search for technical terms and not have Kagi actively change URLs away from my preferred language (which it knows!) just because I live in Spain. Simultaneously, I would really like to search for information and get the locally relevant version, in whatever language that might be. It's tricky, but this is possible in most other search engines today and I'd love to see it in Kagi too.

« Previous Page Next Page »