for some languages, the adjective must be after the noun

[nvaccessAuto]

Reported by drein on 2011-11-24 19:03
For some languages, for example Italian but I'm quite sure also French and Spanish, we put the name and then the adjective.
There are many situations in which NVDA use the English rules, and this is ok, but since now the screen reader is really good, I think we should try to emprove also languages. For example, In a web page, we say "link visited", not "visited link".
It is quite strange to listen "visited link"! Try to imagine in English to put always first the name "link", and then the adjective, "visited".
I think I can't fix this in the .po file.

[nvaccessAuto]

Comment 1 by jteh on 2011-11-24 21:55
Note that not all screen reader speech is entirely natural even in English, largely to maintain consistency in order of speech. For example, we say checkbox checked, but in English, it probably makes more sense to say checked checkbox. This is not something we plan to change even for English. That said, visited link is a hard-coded exception in the code, so this one should probably be made localisable.

[bhavyashah]

@jcsteh Is this localization request something that can be fixed quickly? If yes, please triage and add the QuickFix label accordingly.

[jcsteh]

This probably isn't hugely difficult, but it's not really trivial either.

[Adriani90]

This is quite difficult to achieve for every language. First, you have to know which are all patterns to be changed. Then, you must know the rules of that specific language. There is also checked and not checked, expanded or colapsed, pressed or not pressed, and so on. I don't know how problematic this issue is for other people, but in my opinion it is a very small change which requires much work.

@jcsteh could you give a short guidance how to achieve this for an example language? Maybe we can then write a tutorial on Github and the translators can do it by themselves.

[Adriani90]

in this way the work is splited between translators and it can be done very fast.

[josephsl]

Hi,

More than six years later...

In case of "visited link", this is a combination of control role and state. As Jamie and Adriani (and I'm sure other translators) pointed out, this is a hard task as we need to look for various combinations of role+state to get it right. Also, as Jamie pointed out in a different issue, sometimes screen reader speech is not really grammatically correct in English as well - the job of speech processing part of a screen reader is to analyze incoming text/role/state information and construct speech sequences that makes sense and is consistent across roles and states.

This is one of those cases where we need to remind users to use heuristics to understand text sequences generated by screen readers. I understand the need for screen readers to "say" grammatically correct things across languages (I worked on NVDA translations before so I understand this part). However, a screen reader is a software following exacting (or sometimes heuristic) rules (we call this algorithms), thus we should use it as a reminder to users to practice heuristics at times. As such, I don't think this issue can be resolved - this is a domain for users (humans), not really machines (after all, a screen reader is a machine/processor).

It might be possible that, in a not so distant future, natural language processing (NLP) could be deployed to make screen reader messages more grammatically correct, offering a path to resolve this and similar issues. However, we run into several problems, two of which are:

1. Resource usage: remember that NLP running on a local system can create burden in terms of resource usage, chiefly memory to store language models (more memory means improved accuracy for NLP and other algorithms) and processor burden (mostly central processing unit (CPU), graphics processing unit (GPU), and/or neural processing unit (NPU)). In turn, this translates to increased power usage and system requirements for screen readers (at the moment some developers are talking about resource usage of software, including wattage).
2. Continued need to refine language models: while some languages show remarkable accuracy when processed through NLP, there is work to do to account for more languages to be described/analyzed/modeled/processed/tested/refined at the machine level.

But these problems do not address a more fundamental issue: the purpose of a screen reader and algorithms and data deployed to realize its purpose: announcing text on screen for improved information accessibility. Given its purpose, I think it is more important to get something announced even if it might be inaccurate at times, with human interpretation and heuristics compensating for information the screen reader may not be good at. Therefore, to answer Adriani's point, it is not just translators who must work with screen reading technology - no, it is actually humans who must work with a screen reader to access information; in short, I think it is impossible to rely 100% on screen reader text sequences to get the information the user needs.

What I wrote above could be my (a contributor's) way of saying "no" to this issue. But as I also explained, the situation presented is quite complex and require some thought to unpack, including looking at screen reader text sequence construction process.

Thanks.

[Adriani90]

I think this is work for translators and not something NVDA can change for every language. Closing as can't fix.