"Foreign" language keyword research

[Denyse]

We all know how important keywords (or phrases) are to SEO, but if you work in a language other than english, doing keyword use research can be challenging at best.

Obviously I am more concerned about French phrases, but it applies to any non-english (I am not foreign, to me you are foreign) language in this US/anglophone centric web. Wordtracker only tracks words from anglophone metacrawlers (are there any French metacrawlers?) - yes there are a few french words that show up in there, but they are certainly not representative. Overture has a french database, but I am not sure how representative it is, and not the most friendly to use for large research. There results are based on paid adverts, and certainly the numbers are appalingly low on any search I make.

I know of a few other keyword research aps out there, but I have not found one yet that is professional, and that I would be willing to stake my reputation on.

So my question is, how do you deal with keyword use research in a "foreign language"?

[wiseMouse]

I expect this to be included in localized sites, such as google.fr ... maybe it's a matter of time (prioritizing the US) or mere ignorance...

[Kal]

So my question is, how do you deal with keyword use research in a "foreign language"? It's difficult. My clients that have foreign language versions of their sites usually have an English language site as well so I do the KW research based on that and then either get the client's own native-speaking staff in regional areas to do the translating or use Babel Fish and hope for the best :eek:. Until developers come up with a KW tool for foreign language engines (can you say Market Opportunity?) I think this area will remain hit and miss.

[dannysullivan]

This directory category (http://www.dwoz.com/default.asp?Pr=123) is a nice guide to language-specific keyword research tools offered by Google, Overture and Espotting.

[doppelganger]

I think nothing can replace a good, local human resource... For the nuances of language. I use local resources as a system of checks and balances for all the keyword research I've done with tools.

[Nacho]

The regular version of the Overture Search Term Suggestion Tool (http://inventory.overture.com/d/searchinventory/suggestion/) will cover the U.S. Hispanic market for Spanish terms.

If you're looking for the Spain market, this is the one (http://inventory.overture.com/d/searchinventory/suggestion/?mkt=es) you want to use.

You can also use the Google AdWords Keyword Suggestions (https://adwords.google.com/select/main?cmd=KeywordSandbox) and select language = Spanish and Country = Choose one here (http://forums.searchenginewatch.com/forum/showthread.php?p=3438#post3438)

Otherwise, for Mexico and other Latin American countries, just as Kal said:

Until developers come up with a KW tool for foreign language engines (can you say Market Opportunity?) I think this area will remain hit and miss.
Hopefully Overture will make these markets a priority soon. :)

[steve]

I have to agree with doppleganger. Tools are handy for quick, rough checks, but we do our keyword research through client consultation, competitor analysis, and search engine results analysis. As a rule, we use staff on the ground in each market to ensure local market relevancy and the use of current, correct terminology. Takes more time, but yields far better results.

I think nothing can replace a good, local human resource... For the nuances of language. I use local resources as a system of checks and balances for all the keyword research I've done with tools.

[webcertain]

As a rule, we use staff on the ground in each market to ensure local market relevancy and the use of current, correct terminology.
The problem with this method is that you need to make sure the staff menbers are not using in-house jargon.. It is very difficult for a professional to guess which keyword the general public might use to find their product, that's why wordtracker is such an important tool, and that's why I'm struggling with biased data from Overture and espotting.. I wish one of the main se would accept to sell its datas.. maybe just a comparison tool between keywords..

[lisarenae0561]

I just purchased a keywood tool that allows you to select which country-specific Google search engine and which Google datacenter you want to analyze keywords for. It's called Keyword Elite. I haven't used it all that much yet, but it sounds like just what you are looking for.

Lisa

[webcertain]

Hi Lisa,
thank you for your suggestion.
from what I've gathered keyword elite is only analyzing information from ppc search engine (ie adwords and not google..)which may be helpful for your ppc strategy and kw expansion but may not be the reliable data I'm still looking for....
Cheers,
Johann

[Sebastien Billard]

Well I tend to favor Google Adwords suggestion tool, though any result of this kind of tools should be taken with "a pinch of salt" : they are influenced by ranking checking softwares. Dont forget your common sense then.

Also I do keyword research by frequenting the target audience. Ie if I have to search make-up related keywords, I will go in female forums and communities to have an idea of the real keywords they use.

[Denyse]

Gosh june 2004 seems like such a long time ago. I was just starting off and thanks to the generous responses here, I have learned a lot.

I have tried (and still use) the tools you pointed me to, and since then I also discovered a neat little tool called Keywords Analyser - it does a credible job with french terms (you can choose the data centers and the languages you want), it is not expensive and does a very good job of sending me on different exploration paths.

I agree that there is no substitute for field research because since it seems to gets most of its results from .fr, in Quebec alot of the results are not reliable (almost a different language :D ) so nothing beats your own good judgement.

Of course like with all these tools (Wordtracker included) the results have to be a guide, not the grail.

[Andy AtkinsKruger]

I did a study on this - which has been presented at SES conferences. You can check the relative importance of one phrase over another by comparing the frequency of that phrase in the language.

In other words, you can use the 'results' number on Google to compare one phrase's likely usage against another.

Doesn't give you a count - but when you think about it - counts are nice - but the most important thing to know is whether "triangular widgets" is more likely to appear than "triangle widgets".

Once you've made your list of all the potential terms you think might be suitable - you can use this method to check which is likely to be the most popular. This is most useful in languages where there isn't a readily available tool.

But you still need a native-speaker of the language - which is why we have an office full!!! :)

[orion]

Regarding counts, I can only say this:

Back in 2003 Olsen, Kai A., Williams, James G. published in JASIST Spelling and grammar checking using the Web as a text repository (http://www.asis.org/Bulletin/Oct-04/what_s_new.html). Journal of the American Society for Information Science and Technology, 55(11), 1020-1023.

They suggested that search counts from double quoted queries "could be used to help resolve language decisions such as choice of words, expressions and word relationships for native as well as non-native users of a language." While their paper focused on how Google counts could be used to identify grammatically correct phrases in English, the paper thesis is relevant to any language.


This thesis was highly criticized and was under fire by many linguistics and bloggers. Bob Pritchett (http://bobpritchett.com/blog/posts/298.aspx), for instance wrote:

"Olsen and Williams use Google to show that “stayed in a hotel” is twice as common a usage as “stayed at a hotel”, and they discuss the possible implications of implementing software tools using their methodology. (More confidence for non-native speakers and writers; more conformism in writing.)"

"Is this harnessing the wisdom of crowds? Is it giving license to poor spelling and grammar, because “everyone does it?”* Does it herald the death of style? Or all of the above?"

"* Or should I say “everbody does it?” Everyone edges out everybody 74,900 to 66,600."


This prompted others to make a distinction between prescriptive and descriptive grammar, which Olsen and Williams failed to address in their paper. At the end their thesis was found to be mere speculation.


From an information retrieval standpoint using Google, the thesis as Olsen and Williams stated that "word counts are like a vote of correctness for a phrase" (I'm paraphrasing their claims in the paper) is incorrect and cannot be sustained for the following reasons.


In Google, a search in EXACT mode (double quoted) for "k1 k2" will match documents containing the sequence

"k1 X k2"


where

X = a space, as in phrases.

X = a period, as in ending one sentence with k1 and a period and starting the next one with k2.

X = other delimiters, like "|", ";", " - ", etc.

X = a stop word, like "a", "of", etc.


Therefore, EXACT (double quoted) searches in Google are not necessarily a search for phrases. Consequently Olsen and Williams thesis in the sense that search result counts in Google can be taken for a vote of "phrase correctness" for the initial query is incorrect. (This also applies to searches in other languages).

Such total counts produce a response already contaminated with noise. At best they could be use to have a preliminary idea, only.


This is another reason of why I have argued in the past against the use of absolute search result counts from search engines or keyword discovery tools as they don't necessarily tell the whole story. One can do better by using relative count measures, in order to correct or at least minimize for the noise.


More arguments against the use of absolute counts and keyword discovery tools that overlook the above or use search volume fusion are given in the paper Temporal Co-Occurrence: How does a Developing Event Affects Search Results? (http://www.miislita.com/semantics/c-index-5-temporal-co-occurrence.html)



Orion

[Andy AtkinsKruger]

Orion, I bow to your superior knowledge in this and your paper on EF ratios is fascinating.

I think one of the things we tend to forget when we're looking at search terms is that - whatever method we use - we actually use the past to try and predict the future - and it never turns out quite as we expected.

Your paper puts that point across eloquently - showing how time can change the meaning of a query - such as 'hurricane rita'. For most commercial terms, the biggest impact will be seasonal or marketing impacts (such as a TV ad campaign).

There is a lot of mileage for marketers in studying this area in much more detail as it could help with developing greater understanding of customers and creating new products which satisfy consumer needs - and much more research is needed in this field.

For the search marketer - the essential is to understand that:-


search term counts are always rough estimates
it's better to aim to use the language of the searcher - albeit difficult
search term research needs to be followed by careful on-site performance analysis and adjustments made for customers


And finally, data, data and more data - doesn't necessarily mean information. :)

[orion]

More data counts does not mean more information. That is so true, Andy.

The great thing about the containment measure when looked in time is that allows one to detect changes in trends which are critical to seasonal campaigns and other scenarios.

It is also great for identifying call to actions at the proper time, which one could not do with mere absolute counts. As one passage of the quoted paper says:


"By merely inspecting Figure 2 (absolute count data) one may think that in the websphere and between day 5 and day 15 there was an improvement in search result counts relevant to the query hurricane rita; or that around day 16 and 17 search quality deteriorated. Figure 3 shows this not to be the case. Search results quality actually deteriorated after day 5, the day after landfall. This trend continued for the lifespan of our experiment and can be rationalized from the overall shape of the EF-Ratio curve."


Thus, the metric is great for monitoring document volume or search volume changes for terms highly sensitive to the in-progress event in question. In this case we have the following event-terms to map over time

"hurricane rita" == home insurance, real estate, flooding, first aid kits, etc

No doubt temporal co-occurrence measures can help with calls to action, bidding, etc. in a SEM or marketing campaign.

Back to word usage, similar studies can be done with words in foreign languages or in countries in which in-transit events occur (e.g. wars, natural disasters, announcements of things that will take place at a given time in the future like increase of gas prices after day X, an upcoming movie or song release, tv promotions, etc).

Orion

[edwardro]

I just purchased a keywood tool that allows you to select which country-specific Google search engine and which Google datacenter you want to analyze keywords for. It's called Keyword Elite. I haven't used it all that much yet, but it sounds like just what you are looking for.

Lisa

Lisa, I am trying to get a good keyword tool that allows me to research keywords in spanish in markets such as Mexico, Venezuela, Dominican Republic, Argentina and Spain.

How good is Keyword Elite for that purpose?

Thanks in advance.

Edwardro

[NewKidOnTheBlock]

Hi,

Overture has a french database, but I am not sure how representative it is, and not the most friendly to use for large research. There results are based on paid adverts, and certainly the numbers are appalingly low on any search I make.

I think the reason for those extremely low counts in French (and in German) are, that overture gets their data from yahoo's search engine (and a few very small engines?), but not from Google.

In the US yahoo has a market share of roughly 20% or so(?). But
in France and in Germany yahoo's market share is smaller than 5% (Google has a whooping 85-90% or so).

So technically, if an equal number of people searched for 'dog' and for 'chien' (the french word for dog), overture will display a volume 4 times higher for the English keyword.

Does anybody have an idea if my conclusion is right?

(I'm not sure if that would help you with searches in Québec, though ;-( ).