tgt
stringlengths
1
2.76k
src
stringlengths
1
3.03k
lp
stringclasses
15 values
[PERSON11] To je důležitý.
(PERSON11) That is important.
en-cs
Prosím, zapište to sem do dokumentátoru, tohle shrnutí <unintelligible/>, protože to je pro [PERSON2].
Please write this down here into the documenter this <unintelligible/> summary, because that is for [PERSON2].
en-cs
A taky [PERSON2], až budeš mluvit s [PERSON10], no a taky možná [PERSON5].
And also [PERSON2], when you talk to [PERSON10], well and also possibly [PERSON5].
en-cs
Pokud máme tenhle nástroj, je potřeba (odesílat) zvuk víc ASR nebo víc převaděčům <unintelligible/>
If we have this tool, we need to (ship) the audio to multiple ASR or multiple workers <unintelligible/>
en-cs
Samostatně budeme muset (odesílat) zvuk do anglického ASR, samostatně do německého ASR a českého ASR, například v závislosti na <other_noise/>.
We will separately need to (ship) the audio to the English ASR, separately to the German ASR and the Czech ASR, for example depending on the <other_noise/>
en-cs
A taky do tohohle převaděče ID jazyka a pak potřebujeme tyhle výstupy sloučit, a to je nástroj, který zatím nemáme k dispozici.
And also, to this language ID worker and then we need to merge these outputs, and this is tool which we do not have yet.
en-cs
To je nástroj s víc zdroji, který bude sledovat textový výstupy a taky časový značky a bude vysílat – pravděpodobně bude jakoby vytvářet výstup do tří samostatných jazykových kanálů pro použití buď ticha, nebo správného ASR.
That's the multi-source tool, which will be observing the text outputs and also following the time stamps and it would emit the - it would probably like produce the output into three separate language channels for using either silence or the correct ASR.
en-cs
Takže potřebujeme filtr pro ASR, aby byl ASR zticha, pokud je to nesprávný jazyk, a aby byl rozpoznaný text, pokud je to správný jazyk.
So, we need a filter for the ASRs, so that the ASR is silent if it's the wrong language and it is the recognized text if it's the correct language.
en-cs
Tenhle typ nastavení jsem právě vymyslel.
I just made it up this type of setup.
en-cs
Další z nastavení by bylo, že stejný zvuk bude posílaný do ASR a tohohle kontroloru jazyka.
Another of setup would be that same sound is shipped to ASR and this language checker.
en-cs
Tenhle kontrolor jazyka je v podstatě součástí ASR a v případě nesprávného jazyka ASR umlčí.
This language checker is essentially a part of the ASR, silencing the ASR if it's the wrong language.
en-cs
I to je možnost.
That's also an option.
en-cs
Takže, potřebujeme vymyslet, který způsob integrace ID jazyka je pro naše účely nejlepší.
So, we need to figure out which way of integrating the language ID is the best for our purposes.
en-cs
Přemýšlejte proto o tom, jaký jsou naše pajplajny.
So, please keep thinking about this like what are our pipelines.
en-cs
V ideálním případě by podle mého názoru vzniklo nejmíň zmatků v řízení a tak dál, kdyby naše modely MT byly vícejazyčný.
Ideally, I think that the least like management confusion and so on would arise if our MT models were multilingual.
en-cs
Pokud by podporovaly různý zdrojový jazyky a překládaly z jakéhokoli z nich do angličtiny.
If they supported different source languages and they were translating from any of these languages into English.
en-cs
Případně by stačilo, aby dělaly jen kopii, pokud by byla jako vstup zadaná angličtina, takže bychom na začátku měli vícejazyčný vícenásobný ASR.
Possibly doing just a copy, if the English was given as the input, so we would have multi-lingual multiple ASRs at the beginning.
en-cs
Vše půjde do angličtiny a z angličtiny pak do všech jazyků.
All going into English, and then from the English will go into all of the languages.
en-cs
A později, až bude [PERSON12] mít vícejazyčný model ASR, nebude tohle jazykový ID vůbec potřeba.
And later when [PERSON12] would have multi-lingual model of ASR, that language ID would not even be needed.
en-cs
[PERSON3] Jde o to, že pokud provedeme analýzu správněji, jedna věc je možný protichůdný vyhodnocení, který jen naznačuje, že, že omezení mají skutečně vliv na skutečný výstup.
(PERSON3) The thing is, as long as we do the analysis more properly, one thing is the possible adversarial evaluation, just to indicate that that the constraints are actually having an impact on the actual output.
en-cs
To by bylo hezký a jo, možná i udělat analýzu pozornosti.
That that would be nice and yeah possibly the attention analysis.
en-cs
Ale nejsem si jistý, zda nám to může dát stejný odpovědi, v podstatě, zda, zda model dává, nebo nedává pozor na omezení.
But I'm not sure it might give us the same answers, basically, if the if the model attends or does not attend to the constraint.
en-cs
[PERSON4] Musel jsem se podívat na pár příkladů pozornosti, nedělal jsem si žádnou statistiku ani nic jiného, a podívá se to na omezení, a když to překládá omezení, vypadá to jako daný omezení, a myslím, že by bylo užitečný, kdyby systém dělal nějaký chyby, ale ve skutečnosti žádný v tomhle smyslu nedělá.
(PERSON4) I had to look just at a few examples of attention, I did not do any statistics or anything, and it looks at the constraints, and when it translates the constraints, it looks like the constraints given, and I think it would be useful if the system made any errors, but it in fact does not do any in that sense.
en-cs
Dělá některý obecný chyby v překladu, ale ne ty (jevy), který se snažíme vyřešit.
It makes some general translation errors but not the (phenomena) we are trying to solve.
en-cs
Když se vrátíme k automatickému vyhodnocování, problém s neshodou spočívá v tom, že výstup je správně vyskloňovaný, ale kontext se liší.
Well, going back to the automatic evaluation, the problem with mismatch is that the output is correctly inflected, but the context is different.
en-cs
Nejedná se tedy o stejný slovní druh jako v referenci, že?
Therefore, it is not the same word form as in reference, right?
en-cs
[PERSON4] Jo, ale taky jsem zkontroloval, zda jsou kontexty platný překlady, a ve většině případů jsou.
(PERSON4) Yeah, but I have also checked if the contexts are valid translations and in most cases they are.
en-cs
Stejně jako v prvních 100 příkladů, které byly automatickým vyhodnocením označený jako chybné, bylo 91 z nich správně vyskloňované ve správných kontextech.
Like in the first 100 examples that were marked as error by automatic evaluation 91 of them were correctly inflected in correct contexts.
en-cs
[PERSON4] A myslím, že jen dvě nebo tři věty byly správně vyskloňovaný v nesprávném kontextu, jako by byl překlad špatný.
(PERSON4) And I think only two or three sentences were correctly inflected in the incorrect context, like the translation was wrong.
en-cs
A pak jsou některý případy, kdy byl překlad úplně špatný, protože věta byla opravdu špatně a část musí být <unintelligible/>, ale to bylo jen v jednom nebo dvou případech.
And then there are some cases where the translation was totally wrong, because the sentence was really wrong, and the part must be <unintelligible/>, but that was just like one or two cases.
en-cs
[PERSON3] Ale při vyhodnocování se snažíš o shodu forem odchylek, že?
(PERSON3) But you are trying to match the divert forms when you are evaluating it, right?
en-cs
Neděláš žádný –
You do not do any-
en-cs
[PERSON3] – lemmatizace referencí a –
(PERSON3) -lemmatization of the reference and the-
en-cs
[PERSON4] Obojí, obojí, obojí, snažím se o shodu jak povrchových forem, tak lemmatu.
(PERSON4) Both, both, both, I'm matching both the surface forms and lemma.
en-cs
[PERSON3] Aha, aha, jo, to jsou ty dvě skóre, že?
(PERSON3) Oh I see, oh yeah, those are the two scores, right?
en-cs
[PERSON4] Jen skóre dilematu pro (evropský?) není v tabulce, protože tabulka jako by se nevešla na papír, takže <unintelligible/>
(PERSON4) It is just the dilemma score for the (European?) is not in the table because of the table like won't fit into the paper so <unintelligible/>
en-cs
Pokrytí je vždycky asi 97 procent.
Coverage is always like 97 percent.
en-cs
Prostě to generuje správný lemmata, jen –
It just generates the correct lemmas, just-
en-cs
[PERSON3] Aha, jo, dobře, chápu to, chápu to.
(PERSON3) Oh yeah, okay, I get it, I get it.
en-cs
Pokrytí lemmatu tedy v podstatě říká, zda omezení existuje, a rozdíl v pokrytí povrchu naznačuje, že by to mohlo být nesprávně vyskloňovaný, ale tohle ten případ není.
So basically, the lemma coverage says whether the constraint is there, and the surface coverage difference suggests that it might be incorrectly inflected, but that is not the case.
en-cs
[PERSON3] No jo, jen nevím, jestli jste, jestli jste, jestli jste zachytili <unintelligible/>, když jsme mluvili o [PERSON1],
(PERSON3) Oh yeah, just I do not know if you if you, if you catch <unintelligible/> when we were talking about [PERSON1],
en-cs
ale jedna věc je, že, že (rozptyl) je lepší, ale druhá věc je, že lemmata jsou skutečně správně skloňovaný ručně,
but one thing, is, that, that the (blow) is better but the other thing is that the lemmas are actually properly inflected manually,
en-cs
ale to je, to je další, ne, žádný problém, to je vlastně dobře, že je <unintelligible/> správně, jo.
but that is that is another, no, no problem, it is actually good that is <unintelligible/> right, yeah.
en-cs
Tak jsem si říkal, že by to tam mohlo být.
So, I was thinking there might be.
en-cs
Nejsem si ale jistý, zda pro to máme nějaký testovací data, ale mohli bychom si zkusit pohrát s modely pomocí nějakého přenosu stylu?
But I'm not sure whether we have some test data for that, but we might try to play around with the models with some sort of style transfer?
en-cs
Jak víte, můžeme použít omezení a místo něj zkusit použít synonymní omezení a zjistit nebo porovnat, jak, jak, jaký je rozdíl mezi výstupy.
As you know, we can use a constraint and try to use a synonymous constraint instead and see or compare how, how, what is the difference between the outputs.
en-cs
Ale tohle je, jakoby to popisuji jenom vágně, protože sám nemám přesnou představu, jak to udělat, ale mohl by to být zajímavý, nevšední scénář.
But this is, like I'm just describing it vaguely because I myself do not have an exact idea how to do it, but it might have been an interesting, distinct scenario.
en-cs
Nevím, [PERSON1], víš o nějakých datech o přenosu stylu mezi angličtinou a češtinou? Máme něco k dispozici?
I do not know, [PERSON1], do you know about any style transfer data for English Czech? Do we have something?
en-cs
[PERSON4] <unintelligible/>
(PERSON4) <unintelligible/>
en-cs
[PERSON1] Tím přenosem stylu, Dušane, myslíš, že by tam bylo něco, co je ve spisovném jazyce, a ty bys byl cílovou stranou, bylo by to v mluveném jazyce, nebo něco takového?
(PERSON1) By style transfer, Dusan, you mean that there would be something which is in the written language, and you would be the target side, would be in the spoken language or something like that?
en-cs
[PERSON3] No, to je taková věc, jakoby, mám jen omezený znalosti úlohy.
(PERSON3) Well, that is the thing like, I have only a limited knowledge about the task.
en-cs
Ale dovedu si představit, že chceš přepsat větu tak, že ji nenapsal muž, ale místo něho ji napsala žena, nebo, já nevím, jako že můžeš mít vědecký a nevědecký vysvětlení určitých jevů.
But I can imagine that you want to rewrite a sentence that it is not written by male, but it is written by a female instead or, I do not know, like you can have a scientific and unscientific explanation of certain phenomena.
en-cs
Opravdu nejsem tak dobře obeznámený s přesnými soubory dat a s tím, co pokrývají.
I'm not really that familiar with the exact data sets and what they cover.
en-cs
A jde o to myšlenku toho, že styl vět je poměrně vágně definovaný.
And the idea is that the style of the sentences gets quite vaguely defined.
en-cs
Takže si zase nejsem tak jistý.
So again, not so sure.
en-cs
[PERSON1] Máme tenhle korpus transformací vět.
(PERSON1) We have this corpus of sentence transformations.
en-cs
A jedna z věcí, která v něm je a která by mohla být označená jako přenos stylu, je například větší zobecňování vět.
And one thing that is there and could be labelled as style transfer is for example, making the sentences more general.
en-cs
Takže se vynechávají detaily, věta se zjednodušuje a pak zní hovorově.
So, details are omitted, the sentence is simplified, and then making the sentence sound colloquial.
en-cs
No, je to jen parafráze.
Well, just a paraphrase.
en-cs
A nejsem si jistý, zda máme něco jako zkracování věty.
And I'm not sure if we have anything like shortening of the sentence.
en-cs
Ohledně zkracování dělá Matouš nějaký pokusy mezi angličtinou a češtinou, ale nemáme k dispozici žádný referenční data.
For shortening, Matous is doing some experiment with English–Czech, but we do not have any reference data there.
en-cs
Takže nemáme k dispozici žádný soubor dat pro kompresi vět.
So, we do not have any sentence compression dataset.
en-cs
Možná by tedy zobecnění mohlo být zajímavý, ale obávám se, že existuje příliš mnoho různých dobrých zobecnění, takže jediná reference by byla příliš omezená na to, aby vám řekla něco o kvalitě vašeho zobecnění.
So maybe the generalization could be of interest but I'm afraid that there are too many different good generalizations, so the single reference one would be too limited to tell you anything about the quality of your generalization.
en-cs
A nejsem si vědomý žádných podobných genderových transformací, například by se to dalo udělat <unintelligible/> pro češtinu, takže možná správnou osobou by byl Rudolf Rosa, zeptat se ho, jestli někdy nějaký takový soubor dat nevytvořil.
And I'm not aware of any like gender transformations, for example, this could be done <unintelligible/> for Czech, so maybe the right person would be Rudolf Rosa to ask him if he has ever generated any such dataset.
en-cs
Takže se bude generovat nějaký protějšek věty na základě kořene.
So, there will be a root-based generation of some sentence counterpart.
en-cs
[PERSON3] Myslím, že je to vlastně jakoby zajímavá otázka, zda můžeme použít omezení k vynucení tohohle druhu věcí, jako je rod mluvčího v, v překladu.
(PERSON3) I guess that is actually like an interesting question whether we can use the constraints to enforce this kind of the like gender of the speaker in the in the translation.
en-cs
[PERSON1] To je vlastně hodně dobrý nápad.
(PERSON1) That is actually a very good idea.
en-cs
Takže bychom se mohli zaměřit na tuhle oblast a vytvořit určitou dílčí část testovacího souboru[PROJECT1], která by to pokrývala.
So, we could focus on that and create a particular sub-part of [PROJECT1] test set that would cover that.
en-cs
Aha, takže někdy známe pohlaví mluvčího, takže pokud jste... možná vám nasdílím obrazovku a projdeme si to.
Ah so that we sometimes we know the gender of the speaker so if you are... maybe I'll share the screen and browse that.
en-cs
[PERSON8] Nebo jsem to zkusil s Marianem, který se ho právě účastní, a je to... Nevím, o kolik bodů BLEU nižší.
(PERSON8) Or I tried it with Marian just attending it and it's... I don't know how many BLEU points lower.
en-cs
[PERSON2] Protože jakoby způsob, kterým Martin podle mě dělá <unintelligible/> ty svoje překládky tak, že překládá víc vět najednou a pak si vybere jen tu prostřední a takhle postupuje celým dokumentem.
(PERSON2) Because like the way I believe Martin does his <unintelligible/> little translation, so he translates more sentences at once and then picks only the centre one and goes like this for the whole document.
en-cs
Takže kontext je jakoby v jednom <unintelligible/> spolu s větou, kterou vlastně chceme přeložit.
So, the context is like in one <unintelligible/> together with sentence we actually want to translate.
en-cs
[PERSON10] Jo, myslím – nebyl to Dominik, kdo dělal nějaký experimenty s Ivanou?
(PERSON10) Yeah, I think - wasn't it Dominik doing some experiments with Ivana?
en-cs
Myslím, že v minulém roce nebo před dvěma roky pro hodnotu double empty, kde dávali pozor na kontext nebo řetězili kontext se vstupní větou a dělali nějaký druh překladu na úrovni dokumentu.
I think the last year or two years ago for double empty where they were attending the context or concatenating the context with input sentence and doing some sort of document level translation.
en-cs
Ale ta záležitost s tímhle, se řetězením není zcela jistá, zda má stejný účinek, protože my děláme na nastavení s víc kodéry –
But the thing is with this with the concatenation is not completely sure whether it has the same effect as we do multi encoder setting-
en-cs
[PERSON1] Jakože každý je jiný, je to určitě jiný výpočet, jiný vedoucí, ale to, překvapuje mě, že [PERSON8] říká, že to nefunguje – a tečka.
(PERSON1) Like everybody is different, it's definitely a different calculation, different leader, but it's I'm surprised that [PERSON8] says that it doesn't work full stop.
en-cs
[PERSON8] Jo, dobře, tak se omlouvám, možná bych měl být přesnější, že to nefungovalo jakoby tím způsobem, jak jsem to udělal <laugh/>.
(PERSON8) Yeah, okay, so sorry, maybe I should be more correct that it didn't work like in the way I did it <laugh/>
en-cs
[PERSON1] Jo, protože si myslím, že je to metoda, která se dá velmi snadno testovat, a vždycky by se měla testovat v kontrastu s dvěma kodéry, a očekával bych, že někdy bude lepší a někdy horší než dva nastavené kodéry, v závislosti na přesné úloze, jako co přesně je ta druhá věc, kterou kódujete.
(PERSON1) Yeah, because I think it's a method which is very easy to test, and it always should be tested in contrast with the two encoders, and I would expect sometimes to be better and sometimes to be worse than the two encoders set up, depending on the exact task like what exactly is the second thing that you are encoding.
en-cs
Takže pokud máte dvě kopie dvou parafrází téže věty,
So, maybe if you have two copies two paraphrases of the same sentence,
en-cs
možná je to nějak matoucí z hlediska pozornosti, takže, cokoliv, nevím.
maybe it is somehow confusing for the attention so whatever, I don't know.
en-cs
Chování se taky mohlo lišit u starých metod sekvence po sekvenci ve srovnání s transformátorem, takže u RNN se to pak chovalo jinak, možná u transformátoru je to tak, jo, je těžší ho trénovat, takže možná je třeba zvýšit počet zahřívacích, zahřívacích sad.
Behaviour also could be different for the old sequence-to-sequence methods, compared to transformer, so for RNN then it behaved differently maybe with transformer it's yeah, it's harder to train so maybe you need to whatever increase warm-ups number, warm-up sets.
en-cs
[PERSON2] Jo, takže prostě víš, jakoby –
(PERSON2) Yeah, so just know like-
en-cs
[PERSON8] Nepamatuji si to, takže si to nepamatuji přesně, ale je to jakoby –
(PERSON8) I don't remember, so I don't remember exactly but this is like what-
en-cs
když jsem v létě začal s experimenty s více zdroji.
when I started in summer with those multi-source experiments.
en-cs
Nejdřív jsem začal se řetězením parafráze a zdroje a možná jsem, teď si nevzpomínám,
The first thing I started with was concatenating the paraphrase and source and maybe I, I don't remember now,
en-cs
do příštího mítinku můžu zkontrolovat, zda jsem se o to taky pokusil,
I can check it till the next meeting whether I tried also,
en-cs
určitě jsem zkoušel, když jsou zdroje na první pozici, pak nějaký oddělovač a parafráze na druhé pozici.
I definitely tried when the sources are on the first position then some delimiter and the paraphrase on the second position.
en-cs
Možná jsem taky zkoušel, že jsem ty dvě věci náhodně zamíchal, jo, a jen si vzpomínám, že to hojně dělalo tyhle kompenetrační věci a zkoušelo to víc kodérů.
I might have also tried that I shuffled the two things randomly, yeah, and I just remember that it did abundantly this compenetration stuff and tried multi-encoders.
en-cs
[PERSON8] A to pravděpodobně kvůli nízkému skóre.
(PERSON8) And it was probably due to low scores.
en-cs
[PERSON2] <unintelligible/>, který se ti líbí, ti může zlepšit skóre amenity pouhým připojením výstupu založeného na frázi a pozornost je jako dvojitá diagonála, takže pozornost není problém, takže jakoby věřím ti, ale je překvapující, že to nefungovalo.
(PERSON2) <unintelligible/> that you like can improve your amenity score by just appending phrase-based output and the attention is like double diagonal, so the attention is not an issue, so like I believe you but it's surprising that it didn't work.
en-cs
[PERSON10] Takže pokud zadáte výstup založený na frázi, je to v podstatě posteditování, že? Můžete o tom přemýšlet jako o posteditování, jako byste posteditovali výstup založený na frázi, že? Nebo jakoby záleží na úhlu pohledu, že?
(PERSON10) So, if you if you input the phrase-based output that's basically post editing, right? You can think about it as a postediting like you are postediting phrase-based output, right? Or like depends on the point of view, right?
en-cs
[PERSON1] <unintelligible/>, který se ti líbí plně přepsat, ale technicky přesně odpovídá úloze posteditování, můžeš to zapojit jako řešení úlohy posteditování.
(PERSON1) <unintelligible/> that you like fully to rewrite it but technically it fits exactly the postediting task, you can plug this as a solution to the postediting task.
en-cs
[PERSON10] Dobře, takže jedna otázka jenom za mě, abych se ujistil, že v nastavení řetězení vkládáte token oddělovače vět, že?
(PERSON10) Okay, so one question just for me just to make sure in the concatenation settings you insert the sentence separator token, right?
en-cs
Máte nějaký speciální token, který rozlišuje, která věta je zdrojová a která kontextová nebo jiná, nebo je prostě zřetězíte bez čehokoli?
You have a special token to distinguish like which is the source and which is the context or the other sentence or do you just concatenate them without anything?
en-cs
A doufáte, že se to systém naučí?
And hope that the system learns it?
en-cs
[PERSON8] Zkoušel jsem, tuším, dva tokeny a jo, jakoby protože v první verzi jsme měli podezření, že by ten token mohl být taky tokenizovaný sám o sobě, tak jsem to pak přehrál s jiným tokenem, který není určený jenom pro tenhle účel, protože jsme použili nějaký předtrénovaný slovník, a tenhle slovník neobsahoval tokeny oddělovačů.
(PERSON8) I tried to, I guess two tokens and yeah like because in the first version, we had a suspicion that the token could be also tokenized itself, so then I replayed it with another token which is not only for this purpose because we used some pretrained dictionary, and this dictionary didn't contain separator tokens.
en-cs
Takže jsem použil nějaký token, u kterého jsem předpokládal, že nebude jakoby tokenizovaný na několik částí, a tak se objeví jednou.
So, I used some token that I assumed that it wouldn't be like tokenized into several pieces and so it will appear once.
en-cs
Neměl jsem žádnou záruku, že by se ten token nemohl objevit na jiných místech ve větě a že je to to jediný místo, kde se může objevit.
I had no guarantee that this token couldn't appear in other places in the sentence and that is the only place it could appear.
en-cs
Takže by se to dalo udělat pořádněji, jo, souhlasím.
So, it could be done more properly, yeah, I agree.
en-cs
[PERSON2] Ale část Flask nebo část komunikace s klientem je něco, co nemůžu udělat.
(PERSON2) But the Flask part or the communication part with the client is something that I cannot do.
en-cs