menu
Tatoeba
language
Register Log in
language ქართული
menu
Tatoeba

chevron_right Register

chevron_right Log in

Browse

chevron_right Show random sentence

chevron_right Browse by language

chevron_right Browse by list

chevron_right Browse by tag

chevron_right Browse audio

Community

chevron_right Wall

chevron_right List of all members

chevron_right Languages of members

chevron_right Native speakers

search
clear
swap_horiz
search
FeuDRenais {{ icon }} keyboard_arrow_right

Profile

keyboard_arrow_right

Sentences

keyboard_arrow_right

Vocabulary

keyboard_arrow_right

Reviews

keyboard_arrow_right

Lists

keyboard_arrow_right

Favorites

keyboard_arrow_right

Comments

keyboard_arrow_right

Comments on FeuDRenais's sentences

keyboard_arrow_right

Wall messages

keyboard_arrow_right

Logs

keyboard_arrow_right

Audio

keyboard_arrow_right

Transcriptions

translate

Translate FeuDRenais's sentences

FeuDRenais's messages on the Wall (total 401)

FeuDRenais FeuDRenais 15 ივნისი, 2011 15 ივნისი, 2011, 05:01:27 UTC link Permalink

Tatoeba should have birthday notifications.

That is all.

FeuDRenais FeuDRenais 13 ივნისი, 2011 13 ივნისი, 2011, 17:30:43 UTC link Permalink

That's true too. That's probably the worst case.

FeuDRenais FeuDRenais 13 ივნისი, 2011 13 ივნისი, 2011, 17:19:07 UTC link Permalink

I'm not so sure. Quality gets diluted as you get more people, and currently this site is small enough to allow good monitoring because all the active users take time to leave comments, etc.

If the site is to have major growth, I don't think that the system will be able to retain quality by peer-to-peer comment-based feedback.

But whether or not this site will grow is another question.

FeuDRenais FeuDRenais 13 ივნისი, 2011 13 ივნისი, 2011, 17:03:53 UTC link Permalink

> Actually, the system is quite smart since it improves itself automatically. When people repeatedly create unnatural translations, they get repeated comments, and at some point they get annoyed and realise that their level is not sufficient, and sometimes they leave the project, or sometimes they switch to creating sentences in their native language or correcting mistakes of others and that is fine.


That's not true for everyone.

FeuDRenais FeuDRenais 13 ივნისი, 2011 13 ივნისი, 2011, 16:59:31 UTC link Permalink

Again I would propose having ratings for users. It's unrealistic to expect all the million sentences to be trustworthy, but there need to be guidelines as to which can be trusted.

FeuDRenais FeuDRenais 12 ივნისი, 2011 12 ივნისი, 2011, 16:00:22 UTC link Permalink

> if i write "h a y" in english there's a lot of chance that these are the first letters of "how are you"

Taking a page from Google Pinyin there.

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 18:11:45 UTC link Permalink

Je ne suis pas sûr si on parle toujours de la même chose, mais voilà mon avis honnête:

Tatoeba est splendide, mais elle manque un algorithme pour diriger automatiquement le grand nombre de données qu'elle reçoit quotidiennement, et pour dessiner la ligne entre le bon et le mal.

C'est bien d'avoir des indications et l'unité de la communauté pour encourager le comportement "correct", mais je crois que c'est désespère à la longue, vu, encore, le grand nombre de phrases qu'on collectionne, ainsi que la diversité des contributeurs.

Mais bon, ce n'est que mon avis.

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 16:54:13 UTC link Permalink

On peut essayer de vérifier une chose: est-ce que le nombre d'étiquettes NNC croise par rapport au nombre de phrases totale ?

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 16:51:18 UTC link Permalink

Cependant, ce n'est pas automatique. Ça demande qu'on consacre le temps (peut être que t'en as, mais je pense pas que c'est vrai en général). En plus, on peut pas mettre chaque nouveau contributeur en quarantaine - c'est offensant et souvent pas nécessaire. Enfin, ça n'arrête pas un contributeur expérimenté qui décide, un jour, d'essayer d'ajouter des phrases dans une langue qu'il est en train d'apprendre.

Ce que je propose laisse les gens qui veulent traduire pour apprendre traduire pour apprendre. Mais au lieu de masquer ça, ou de naviguer la zone grise qui correspond à "est-ce que c'est naturel ou pas, faut-il que je mets un NNC ici?", on peut traduire sans contrainte (ils vont être, néanmoins, pénalisés si les traductions ne sont pas bonnes, et le score correspondra à ça).

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 16:16:31 UTC link Permalink

Et non, je ne m'intéresse pas non plus au score personnel. C'est juste une façon de régler tant de problèmes à la fois.

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 16:14:16 UTC link Permalink

Oui, mais il faut constater qu'il y a à peu près un million de phrases sur Tatoeba, et seulement quelques milles de contributeurs (dont une fraction est responsable pour une grande partie de phrases).

Si tu attaques directement les phrases, tu auras trop de travail à faire, car tu n'attaques pas la source. Disons qu'un nouveau (mauvais) contributeur ajoute tout à coup 100 phrases qui sont soit liées de manière erronée, soit n'ont pas du sens, soit sont mal traduites. Pour réparer tout ça, il faut un grand effort de notre part (juste pour un seul cas). C'est à cause de ça que, par exemple, sysko a limité la liberté de boracasli.

Si tu attaques la source (le contributeur), tu rends inoffensif tout ce qu'il fait désormais, même s'il continue à ajouter des phrases, voire un mille d'autre phrases qui ont des problèmes ou qui sont erronées. C'est beaucoup plus pratique que d'essayer de régler les problèmes phrase-par-phrase.

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 15:31:52 UTC link Permalink

Also, you have to realize that a rating system is probably the only way to give this website any professional credibility. I mean, how else do you make guarantees about quality if not via feedback/ratings? Trang can't just say "you can trust sacredceltic's translations - believe us on this". It's a much more powerful statement to say that "this user has had his translations rated 10,000 times and has a rating of 98.5%" or something of the sort.

It would also eliminate so many of these other problems - native checks, debates about whether or not non-natives should translate, blocking users who are adding bad translations, orphans, etc. Bad translations would just have bad ratings and no one would trust them - problem solved.

FeuDRenais FeuDRenais 8 ივნისი, 2011 8 ივნისი, 2011, 13:31:37 UTC link Permalink

"une catastrophe" - on peut faire pire avec les traductions qu'avec les liens (et tout le monde peut traduire, sans même connaitre les règles)...

Je suis, encore une fois, confondu par comment certains considèrent les liens comme une espèce d'arme nucléaire qu'il faut contrôler à n'importe quel prix.

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 22:18:56 UTC link Permalink

(I know)

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 22:00:04 UTC link Permalink

Assuming total chaos amongst members, you have a point. But I wouldn't say there's such a chaos here. It's not very hard for experienced members to know who's good for which language pairs. If we can't trust that, then indeed it's complete chaos and no rating system can ever work. But it's not that hopeless, IMO.

To solve the problem you bring up, you'd have weighted rating (i.e. experienced members, or members with higher ratings, would have much more powerful votes than completely new members who haven't been established). All you need to do is initialize these high ratings for a few people that you trust.

e.g. You are given a very high rating for French-English by default because people (we will assume) trust you in this domain. Then you rate other French-English translations, and the ones who get rated well have their ratings improved from 0 to something better than 0. At first, their own votes won't matter by comparison to yours, but over time they grow and the best rated also become the most powerful raters. It's a quality-breeds-quality system (you just have to start with quality).

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 20:58:08 UTC link Permalink

Well, here goes a really reckless idea, but...

1) Delete all the orphans on Tatoeba. It won't feel as bad if we call them "weeds" instead of "orphans" and admit the fact that they're causing a lot of problems and are, in many cases, unreliable.

2) Institute ratings for links only. Since links all have owners, users will be rated for their translations and credibility will be established (i.e. sacredceltic will have a very high rating for French-English, fucongcong for Mandarin-French, etc.). That way, a person who doesn't know Tatoeba but wants to know if a certain translation is good will only have to look to the person responsible for the link, and if the person is well-rated, then the translation is trustworthy.

Quick 2-step solution. Any holes here?

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 20:48:01 UTC link Permalink

Just nuke them all.

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 20:47:29 UTC link Permalink

Not, I mean... ALL of them. Not just your 10,000.

FeuDRenais FeuDRenais 7 ივნისი, 2011 7 ივნისი, 2011, 19:23:25 UTC link Permalink

Hey... why not just delete all the orphan sentences?

Clean slate!

FeuDRenais FeuDRenais 6 ივნისი, 2011 6 ივნისი, 2011, 13:34:06 UTC link Permalink

+1