Our data is released under various Creative Commons licenses.More information
If you love this content, please consider a donation.
Wall (6,716 threads)
Tips
Before asking a question, make sure to read the FAQ.
We aim to maintain a healthy atmosphere for civilized discussions. Please read our rules against bad behavior.
sacredceltic
2 days ago
Ergulis
3 days ago
CK
4 days ago
Rafik
4 days ago
Pfirsichbaeumchen
4 days ago
odexed
4 days ago
small_snow
4 days ago
bill
4 days ago
Cangarejo
4 days ago
alexmarcelo
4 days ago
🍎 A list with over 2,000 Japanese sentences that don't have kanji
https://tatoeba.org/en/sentences_lists/show/170911
All sentences on this list are owned by native Japanese speakers.
Maybe Kanjis are out of fashion nowadays with younger natives…
Belated happy birthday, @Carlosalberto
Félicitations à Carlos qui souffle aujourd'hui ses 90 bougies. 🥧🍾🎁
Boldog kilencvenedik születésnapot!
Isten éltessen sokáig, füled érjen bokáig!
Feliz aniversário Calos.
Parabéns! ¡Que te las cumplas feliz!
Parabéns ao querido companheiro!
Parabéns!
Parabéns pelos seus 90 anos, @carlosalberto!
@carlosalberto, Happy 90th birthday! 卆寿
I wish you good health for many years to come.
@carlosalberto поздравляю с юбилеем!
@carlosalberto
Noch einmal meinen allerherzlichsten Glückwunsch zum Geburtstag, lieber Carlos. Mögen uns noch viele Jahre des gemeinsamen Sprachenlernens bevorstehen! 🙂
Amulli ameggaz @Carlosalberto. 🍾🎂🎊
Happy birthday! I wish you all the best .
Carlosalberto, você precisa atualizar o texto em seu perfil.
Muitos parabéns!
Parabéns, Carlos!
✹✹ Stats & Graphs ✹✹
Tatoeba Stats, Graphs & Charts have been updated:
https://tatoeba.j-langtools.com/allstats/
🍎 2023-01-21 Stats Based on http://bit.ly/nativespeakers
language : sentences by native speakers : native speakers
eng : 1081592 : 1532
por : 395655 : 676
rus : 920306 : 561
fra : 418275 : 519
tur : 272479 : 484
spa : 308209 : 479
deu : 541846 : 476
pol : 110110 : 296
hun : 369428 : 210
jpn : 113500 : 200
ita : 808651 : 185
kab : 598351 : 178
nld : 120746 : 138
fin : 139143 : 79
ber : 612414 : 74
ukr : 167646 : 68
heb : 193424 : 51
cmn : 49185 : 226
ara : 29396 : 190
swe : 44081 : 75
pes : 25017 : 74
vie : 20975 : 69
ces : 58210 : 68
ind : 16723 : 65
ron : 24931 : 47
ell : 28680 : 36
nob : 11579 : 36
dan : 54539 : 32
bul : 23723 : 28
tgl : 11701 : 27
srp : 45759 : 22
lit : 75851 : 15
slk : 14785 : 14
bel : 10721 : 12
isl : 10763 : 10
mkd : 77908 : 7
eus : 2894 : 72
tat : 10049 : 6
kor : 9108 : 50
mar : 73153 : 4
cat : 6088 : 35
tuk : 6381 : 30
hrv : 4452 : 24
hin : 4302 : 21
yue : 5320 : 18
est : 4395 : 18
tha : 4474 : 16
afr : 2475 : 15
aze : 4188 : 14
urd : 1633 : 12
nds : 15987 : 1
lvs : 2182 : 11
kmr : 6009 : 9
zsm : 1549 : 9
glg : 3267 : 8
nus : 2034 : 8
sqi : 2002 : 7
bre : 6395 : 6
ben : 5347 : 6
uig : 2952 : 6
chv : 1422 : 5
ilo : 2458 : 4
kha : 2085 : 4
zza : 1621 : 4
asm : 3675 : 3
slv : 736 : 28
oss : 3001 : 2
cbk : 2607 : 2
war : 2019 : 2
khm : 1475 : 2
mhr : 3034 : 1
dtp : 2586 : 1
pam : 1480 : 1
kat : 310 : 8
mal : 858 : 6
ceb : 834 : 6
ckb : 366 : 6
bos : 160 : 6
mon : 100 : 6
mya : 538 : 5
arq : 319 : 5
ota : 950 : 4
tam : 275 : 4
shy : 209 : 4
tel : 196 : 4
bal : 140 : 4
kaz : 131 : 4
ori : 367 : 3
kan : 181 : 3
rom : 858 : 2
hoc : 519 : 2
cym : 335 : 2
run : 280 : 2
uzb : 130 : 2
swh : 118 : 2
xmf : 775 : 1
frr : 555 : 1
max : 406 : 1
xal : 377 : 1
nst : 362 : 1
mfa : 319 : 1
mri : 304 : 1
hye : 180 : 1
vec : 165 : 1
guj : 156 : 1
cha : 154 : 1
krc : 139 : 1
shi : 135 : 1
fao : 129 : 1
rue : 116 : 1
afb : 96 : 5
sdh : 53 : 4
zgh : 48 : 4
cjy : 13 : 4
sat : 76 : 3
arz : 68 : 3
pus : 29 : 3
ibo : 20 : 3
ary : 18 : 3
wol : 11 : 3
acm : 47 : 2
som : 42 : 2
ast : 40 : 2
kir : 38 : 2
avk : 36 : 2
akl : 28 : 2
gsw : 26 : 2
hau : 15 : 2
pms : 87 : 1
sah : 84 : 1
ext : 65 : 1
oci : 62 : 1
kbd : 61 : 1
sgs : 46 : 1
tet : 46 : 1
pag : 41 : 1
div : 33 : 1
bho : 32 : 1
tah : 28 : 1
jam : 26 : 1
niu : 25 : 1
aoz : 24 : 1
que : 23 : 1
pap : 20 : 1
tir : 20 : 1
mlt : 19 : 1
mlg : 15 : 1
sun : 15 : 1
toi : 14 : 1
tvl : 11 : 1
sin : 7 : 3
npi : 8 : 2
ltz : 7 : 2
bak : 6 : 2
brx : 5 : 2
tgk : 2 : 2
grc : 9 : 1
mic : 8 : 1
che : 6 : 1
apc : 5 : 1
gom : 4 : 1
kjh : 4 : 1
aii : 3 : 1
amh : 3 : 1
mni : 3 : 1
yor : 3 : 1
zlm : 3 : 1
bar : 2 : 1
bcl : 2 : 1
ksh : 1 : 1
nno : 1 : 1
rif : 1 : 1
🍎 Tatoeba.org Native Speakers with Native Language Sentences
http://bit.ly/nativespeakers
Find native speakers of languages you are studying and get links to their native language sentences.
Updated: 2023-01-21
To continue the discussion from these threads:
https://tatoeba.org/en/wall/sho...#message_39333
https://tatoeba.org/en/wall/sho...#message_39361
https://tatoeba.org/en/wall/sho...#message_39380
https://tatoeba.org/en/wall/sho...#message_39395
Humans need variety and context in order to understand and learn language. A single word may have several meanings. It may be used with some words but not others. It may have a level of formality that makes it only suitable for certain situations. If these nuances are to be induced from a collection of sentences, the collection must be sufficiently diverse. In addition, the human mind needs constant stimulation. Duplicate or near-duplicate sentences create boredom, and a bored mind refuses to learn.
The set of sentences "They learned some Greek while they were on vacation", "Rima and Skura learned valuable life lessons the hard way", and "Our mother, we learned later, had been saving money in a separate account" demonstrate different kinds of things that can be learned (in addition to teaching additional vocabulary). By contrast, the set of sentences "<X> learned Berber", "<Y> learned Berber", and "<Z> learned Berber" tells you nothing about Berber other than that it can be learned, teaches you nothing about learning other than that it is a process that can be applied to Berber, and gives you no vivid picture that might help your mind retain the words or even simply brighten your day. If X, Y, and Z have different gender, number, and person, and the sentences are translated into a language where this affects the form of the equivalent of the word "learned", this set of sentences can show you how to conjugate the verb "learn" in the past tense, but it won't tell you anything else. It won't even tell you as much about verb conjugation as a verb chart will. A set of sentences in random order can't give you visual clues, such as placement on the page, that bring out the inherent patterns. And with no starting point, there's no suitable place to explain important points such as where the patterns hold and where they change or break down.
Sixteen years ago, there was no shortage of dictionaries, or verb charts, or descriptions of other aspects of grammar, either printed or in electronic form. But it was hard to find to find searchable collections of sample sentences that would demonstrate the usage of words and phrases. Trang founded what became Tatoeba in order to address her own frustration with that shortage. She opened her new project up to accept contributions from virtually anyone, with no barrier to entry other than the ability to work on an electronic device over the Internet using a free and anonymous account, very few restrictions other than a set of gradually evolving guidelines that were minimally enforced, and an interface that enabled correction of errors after they had been introduced instead of a moderated pipeline that would require sentences to be approved before going live. The website also put itself in the world of social media by providing the ability to leave comments on sentences and on a general "wall", and provided statistics on the number of contributions by language and by individual contributor.
This structure was both a blessing and a curse. It contained the potential for a wide variety of people around the world to collaborate in harmony to contribute an almost infinitely diverse set of utterances. It also contained the potential for individuals to antagonize each other while mass-producing colorless sentences of minimal value that were less valuable at Tatoeba than they would have been elsewhere.
Tatoeba is somewhere between these extremes of paradise and hell. There are many worthwhile corners of variety in the corpus for people who know how to look for it and are willing to expend the effort to extricate the jewels from the trash. The problem is that large-scale contribution of near-duplicate sentences makes this harder and harder to do. Since those sentences have similar content and are added to Tatoeba at the same time, they'll all show up in the same place in your results, and will crowd out everything else. I'm not saying that adding ten "<> learned some Berber" sentences is a problem, or that someone needs to panic about adding a sentence that might be similar to one that someone else has already added. But adding a hundred "<> learned some Berber" and a hundred "<> will learn some Berber" sentences, or going on a rampage of adding a "Ziri" copy of every "Tom" sentence you can find, is another matter.
Furthermore, the elimination of variety tends to propagate itself via translation, or via copycat or retaliatory behavior. In an unintentionally ironic post below (https://tatoeba.org/en/wall/sho...essage_39426), CK filled a screen with a near-duplicate series of near-duplicate sentence pairs (35 of them!) in an apparent attempt to bolster his assertion that eliminating variety in the use of proper nouns is a way to avoid lack of variety in other aspects of language. I think the preposterousness of this idea is self-evident: it's like claiming that preventing someone from learning history is going to make them a better math student. But it's especially worth pointing out that many of those duplicate sentences were added in reaction to the ubiquity of the name "Tom" throughout the corpus. So CK, or more broadly speaking, the general elimination of one aspect of variety, had a large hand in creating the problem in the first place.
It would be nice to introduce some mechanized way of mitigating the problems introduced by mechanized creation of sentences (including using one's mind like a machine, whether or not a computer is involved). But as I have already described, I don't think this can be done easily, quickly, or in a way that would approach any meaningful consensus. Instead, I want to explain to the people who are tempted to add near-duplicate sentences to Tatoeba on an industrial scale that they are probably better off doing the same thing elsewhere, and that there are easy ways to do it that they might not have thought of.
Let's say you want to collect sentences and audio containing Kabyle placenames for a GPS, or that you want to cover all the conjugated forms of Berber verbs that you can think of. It's not as though you have thousands of contributors in unison acting on a top-security project that requires millisecond response time. Is it not simply possible to write the sentences in a spreadsheet on your computer and put the audio files you record into a folder? Or in Google Docs or Google Drive? Or set up your own simple website? (A simple search for "easy free ways to set up a database on the web" brings up 924 million results.) They'll be easier to retrieve from such a place anyway. Why involve Tatoeba at all?
Choosing the best site to help you prepare a special project takes a little bit of imagination and a little bit of effort. So does contributing sentences with enough variety to be worth something to others at Tatoeba. Don't let that stop you.
Has an end user ever complained about near duplicates?
Has anyone ever said, "I'm not going to use Tatoeba's corpus because of the number of near duplicates"?
It seems like a huge drain on time and resources to care about them, when there are more serious problems. Near duplicates are not a problem that affects anyone; hate speech is. The worst thing that can result from near duplicates is frustration or boredom; the worst thing that can result from hate speech is genocide.
Más oldalak, melyek felhasználják az itteni mondatokat, mind rendszerezték valamely formában azokat.
Itt nincs rendszerezve, random mondatok közt vagy egy 20-as mondatlista lefordítása közben hamar megunja az ember magát ha csupa Lvl 1 mondatot talál, vagy ugyanolyanokat, csak más szereplővel.
Vagy legyen rendszerezhetőbb itt pár dolog, esetleg precízebb keresési kritériumok, vagy legyen megállapodás arról, hogy kell-e nekünk ennyi adat és nem pedig információ.
Maybe not affects anyone, but more poeple than I first thought.
No, what you find hate speech, also not affects anyone, but probably some, even many, still, it is not that wall message what is about hate speech.
> Itt nincs rendszerezve, random mondatok közt vagy egy 20-as mondatlista lefordítása közben hamar megunja az ember magát ha csupa Lvl 1 mondatot talál, vagy ugyanolyanokat, csak más szereplővel.
Have you tried using Tatominer?
Yes, I did, tried some times, but the problem was the system not recognized correctly the words, the Hungarian language uses a ton of word endings.
There are lemmatizers for Hungarian, but maybe they’re too slow.
Yes, quite sure there have been "end users" put off by the usefulness of the content considering the size. If you have nothing to add, that's fine but please give up on attempting to hijack the topic, at least.
It feels sad but thank you for taking the time to write about these issues. Call me a "technical solutionist", but I believe there are technical ways to at least mitigate the problem, in addition to convincing members to change their behavior. But as you said, these are costly so they need to be carefully thought before implementing.
Just to make sure we are talking about the same thing: Basically, you cannot get useful results when searching because all the results look similar, so you need to scroll over and over in order to find useful sentences.
I saw interesting ideas here:
https://github.com/Tatoeba/tatoeba2/issues/2816
I'd like to suggest the following approaches, in order of estimated cost of development (from cheapest to most expensive):
- add a "number of words" search filter to easily exclude short sentences
- expose per-language (and per-user?) statistics about sentences "diversity", similar to what can be seen in the Github issue
- introduce a new smarter ranking algorithm that favor sentences based on their uniqueness, length or other criterion, using GDEX as an inspiration https://www.sketchengine.eu/guide/gdex/ ). This however brings new political problems of how to decide on the criteria.
- cluster search results so that similar sentences are grouped, only display one sentence of each group, but allow clicking on a group to see all the hidden sentences
@gillux Thank you for the update on these technical solutions. I hope we will find time to implement some of them in the future.
I would like to add to your list of features the possibility to "mute" a contributor in the settings: https://github.com/Tatoeba/tatoeba2/issues/2008
@gillux Now that we've had a chance to hold a comprehensive discussion of behavior, I'm happy to have us discuss technical solutions. Any of the ones that you and @lbdx mentioned would be helpful.
Until we are able to implement these technical approaches, people can use random search when looking up words or when finding candidate sentences to translate. One additional technical solution would be to make random search the default, rather than "relevance" (which favors shorter sentences and tends to show near-duplicates in clusters). I can imagine some resistance to this idea, since some users probably prefer seeing the shorter sentences even if they lack context, but it's something we can discuss.
I believe this is a search issue rather than a contribution issue.
Here's an example.
The word "content" has the following word senses (from Wiktionary):
A. satisfied
B. that which is contained
C. subject matter
D. the amount of material contained
E. mathematics: space contained by a polytope
As a thought experiment, suppose that the corpus contains 100 example sentences
of each word sense. If the search results displayed all the A sentences before all the B sentences, and all the B sentences before all the C sentences, etc., then there's going to be a perceived lack of diversity, regardless of how diverse the corpus really is.
To be able to filter the results according to the senses of the searched word would indeed be a very nice feature. However, I sincerely doubt that we will be able to implement it for more than 400 languages anytime soon.
> If the search results displayed all the A sentences before all the B sentences, and all the B sentences before all the C sentences, etc., then there's going to be a perceived lack of diversity, regardless of how diverse the corpus really is.
Diversity in terms of word sense is a valid thing to consider, but it's the kind of criterion that someone would apply only if they were doing an analysis that required some thought. I was talking about the lack of diversity that jumps out at someone instantly because their search results are all identical except for variation in a single respect (such as <placename> in sentences of the form "I drove to <placename>").
Also, sentences with sense A, B, C, etc. will only end up clustered together in the search results if they share a characteristic with the criterion beng used for the sort (for instance, the user is searching by sentence creation date, and sentences with sense A all happened to be added before sentences with sense B). And sentences with sense A will only crowd sentences with sense B off the page if there are lots of sentences with sense A. For these reasons, I think it's unlikely that there are many pockets of "hidden sense diversity" that can be found in the tail end of search results but are not visible closer to the beginning.
It's true that there may be more diversity in the corpus than is visible on a single page of search results. However, we can't expect all users to either scroll for an indefinitely long period of time or use sorting criteria to increase the variety at the top of the search results.
> I want to explain to the people who are tempted to add near-duplicate sentences to Tatoeba on an industrial scale that they are probably better off doing the same thing elsewhere
@AlanF_US Thank you for taking the time to talk to the few Tatoebans involved. It looks like you've managed to convince them to reduce the volume of their contributions (at least temporarily). This is really good news 🎉
Some of the larger contributors may be wondering if the sentences they have added recently are diverse enough. You can check the table at the bottom of https://colab.research.google.c...P8?usp=sharing that gives a lexical diversity score to the current month's contributions. The measure used is called MTLD. It is recognized as reliable because it does not vary with the length of the texts analyzed. The higher the MTLD score, the more diverse the sentences added by a contributor. If your score is below 25, your sentences probably contain a high proportion of near-duplicates.
> It looks like you've managed to convince them to reduce the volume of their contributions (at least temporarily). This is really good news
I'm happy to hear that.
The idea of displaying the MTLD for a contributor is interesting. Would it be feasible to give someone the ability to do one of the following?
(1) see the figures for the top X contributors, where X is some arbitrary number other than the one you're already using
(2) type in a username to see the figure for that contributor alone
For now, the online tatowatch notebook is read-only but those who would like to edit and run it can copy it to their Google drive or download it to their own machine.
The idea of this project is to help the moderation team to keep an eye on new contributions. If any of you have feature suggestions or want to participate in this project, please feel free to contact me in private message.
Azul,
Seɣtaɣ ugar n 11.000 n tefyar deg kra n wussan kan (s uṭagi aṛbiɛi/adal). Mazal, arma fukkeɣ-tent-akk (≈304.000).
Ttbeddileɣ tidak n ṣṣmaṭa, akken ad lhunt u ad sɛunt lmeɛna/anamek.
Ad yili usuqel ɣer tefransist neɣ taglizit akken ad sɛunt akk assaɣen. Yak d wa i d iswi!
Mazal ad d-rnun iwiziwen/tiwiziwin, ussan-a, akken ad s-nefk ddḥis, axaṭeṛ nxuṣṣ mliḥ deg usuqel.
Afud uzmir i yal yiwet d yal yiwen.
Azul. Walaɣ aṭas i tseggmeḍ, d ayen igerrzen, nessaram ad seggment akk tid i d-terniḍ tineggura-a, ladɣa anda llant talɣiwin ur neddi akken ilaq deg teqbaylit, ulamma ahat d ayen ara k-yeṭṭfen akud, l. Afud iǧehden.
Tanemmirt a Samir,
Ad dhuɣ deg-sent akk, akken ma llant. Yiwet yiwet, awal awal, asekkil asekkil!
Kkes ugur akk!
Afud uzmir ula i kečč.
:-)
Afud igerrzen ay Igider.
Ayen i d-nenna drus, d leqdic i yewwin aṭas seg wakud-inek, am uzal am yiḍ. Ssarameɣ ad yeffeɣ leɛtab-ik ɣer tafat. Tanemmirt.
Tanemmirt a Rafiq,
Ad d-iniɣ ula d nekk akken ɣef leqdic-ik. Akken i k-nesnemmer drus!
Awi-d kan ad teddu teqbaylit (Teqvaylit) ɣer sdat u ad taf iman-is akken ilaq.
Afud uǧhid ula i kečč.
I've added a number of new English audio files to recently-added sentences.
https://tatoeba.org/en/sentence.../show/4000/und
Perhaps you might want to translate some of these into your own native language.