tgt stringlengths 1 2.76k | src stringlengths 1 3.03k | lp stringclasses 15
values |
|---|---|---|
(אדם8) או ניסיתי את זה עם מריאן שרק השתתפה בזה וזה... אני כבר לא יודע כמה נקודות BLEU נמוך יותר. | (PERSON8) Or I tried it with Marian just attending it and it's... I don't know how many BLEU points lower. | en-he |
(אדם2) מפני שהדרך שאני מאמין שמרטין עושה את ה-<לא מובן/> התרגום הקצר שלו, הוא מתרגם כמה משפטים בבת אחת ואז בוחר רק את המשפט המרכזי וכך במסמך כולו. | (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-he |
כך שההקשר הוא כמו ב-<לא מובן/> יחד עם המשפט שאנחנו למעשה רוצים לתרגם. | So, the context is like in one <unintelligible/> together with sentence we actually want to translate. | en-he |
(אדם10), כן, אני חושב – זה לא היה דומיניק שעשה את אותו ניסיון עם איוונה? | (PERSON10) Yeah, I think - wasn't it Dominik doing some experiments with Ivana? | en-he |
אני חושב ששנה שעברה או לפני שנתיים עבור ריק כפול ששם הם רצו את ההקשר או ריכזו את ההקשר עם משפט קלט וביצעו תרגום כלשהו ברמת המסמך. | 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-he |
אבל העניין הוא שלא ברור אם לריכוז יש את אותה הפשעה כמו שאנחנו עושים הגדרה של קידוד מרובה- | 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-he |
(אדם1) כל אחד הוא שונה, זה בפירוש חישוב שונה, מנהיג שונה, אבל אני מופתע ש-[אדם8] אומר שזה לא עובד, נקודה. | (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-he |
(אדם8) כן, אוקיי, סליחה, אולי אני צריך להיות יותר מדויק שזה לא בד בדרך שעשיתי את זה <צחוק/> | (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-he |
(אדם1) כן, כי אני חושב שזו שיטה שקל מאוד לבחון, ותמיד כדאי לבחון אותה בניגוד לשני המקודדים, ואני מצפה שלפעמים זה יהיה יותר טוב ולפעמים יותר גרוע מהגדרה של שני מקודדים, תלוי במשימה המדויקת כמו מה זה בדיוק בדבר השני שאתה מקודד. | (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 enc... | en-he |
אז אולי אם יש שני העתקים, שני שכתובים של אותו משפט, | So, maybe if you have two copies two paraphrases of the same sentence, | en-he |
אולי זה קצת מבלבל מבחינת תשומת הלב או סיבה כלשהי, אני לא יודע. | maybe it is somehow confusing for the attention so whatever, I don't know. | en-he |
ההתנהגות יכולה גם להיות שונה בשיטות רצף-לרצף, בהשוואה לטרנספורמטור אז ב-RNN הוא התנהג שונה, אולי עם הטרנספורמטור זה כן, קשה יותר להכשיר אותו אז אולי צריך להגביר את מספר החימומים, את הסטים של החימומים. | 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-he |
(אדם2) כן, אז כמו- | (PERSON2) Yeah, so just know like- | en-he |
(אדם8) אני לא זוכר, אז אני לא זוכר בדיוק אבל זה בדיוק כמו מה- | (PERSON8) I don't remember, so I don't remember exactly but this is like what- | en-he |
כשהתחלתי בקיץ עם הניסויים האלה מרובי המקורות. | when I started in summer with those multi-source experiments. | en-he |
הדבר הראשון שהתחלתי היה ליצור שרשור של הניסוח מחדש והמקור ואולי, אני לא זוכר כרגע, | The first thing I started with was concatenating the paraphrase and source and maybe I, I don't remember now, | en-he |
אני יכול לבדוק עד הפגישה הבאה אם ניסיתי גם, | I can check it till the next meeting whether I tried also, | en-he |
אני בהחלט ניסיתי כשמקורות היו בעמדה הראשונה ואז איזה תוחם וניסוח מחדש בעמדה השניה. | I definitely tried when the sources are on the first position then some delimiter and the paraphrase on the second position. | en-he |
יכול להיות גם שניסיתי לערבב את שני הדברים באופן רנדומלי, כן, ואני רק זוכר שזה נתן הרבה תמורות וניסה מקודדים מרובים. | 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-he |
(אדם8) וזה היה כנראה בגלל ניקודים נמוכים. | (PERSON8) And it was probably due to low scores. | en-he |
(אדם2) <בלתי מובן/> שאתה רוצה לשפר את ציון הנוחות על ידי הוספת פלט מבוסס-מונח והתשומה היא כמו אלכסון כפול כך שהתשומה היא לא בעיה, אז אני מאמין לך אבל זה מפתיע שזה לא עבד. | (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-he |
(אדם10) אז אם אתה מזין את הפלט מבוסס המונל זה בעצם לאחר עריכה, נכון? אפשר לחשוב על זה כעל לאחר עריכה כמו שאתה עורך פלט מבוסס-מונח, נכון? או שזה תלוי בנקודת המבט, נכון? | (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-he |
(אדם1) <לא מובן/> שאתה צריך לשכתב את זה מחדש אבל מבחינה טכנית זה מתאים בדיוק למשימה שלאחר העריכה, אפשר לשלב את זה כפתרון למשימה שלאחר העריכה. | (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-he |
(אדם10) אוקיי, אז שאלה אחת רק כדי לוודא שבהגדרות של שרשור אתה מזין את סמל מפריד המשפטים, נכון? | (PERSON10) Okay, so one question just for me just to make sure in the concatenation settings you insert the sentence separator token, right? | en-he |
יש לך סמל מיוחד להפרדה בין המקור וההקשר או המשפט האחר או שאתה משרשר אותם ללא דבר? | 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-he |
ומקווה שהמערכת תלמד את זה? | And hope that the system learns it? | en-he |
(אדם8) ניסיתי, אני משער ששני סמלים וכן כי בגרסה הראשונה, חשדנו שהסמל יכול להיות מסומן בעצמו, אז הפעלתי שוב עם סמל אחר שלא נועד רק למטרה הזו כי השתמשנו במילון שהכשרנו מראש, והמילון הזה לא הכיל סמלים מפרידים. | (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 toke... | en-he |
אז השתמשתי בסמל כלשהו שהנחתי שלא יסומן בכמה חלקים כך שהוא יופיע פעם אחת. | 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-he |
לא היה לי ביטחון שהסמל הזה לא ייעלם במקומות אחרים במשפט ושזה המקום היחיד שבו הוא יכול להופיע. | 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-he |
אז אפשר לעשות את זה בצורה נכונה יותר, כן, אני מסכים. | So, it could be done more properly, yeah, I agree. | en-he |
(דמות2) אבל החלק של Flask או התקשורת עם הלקוח זה משהו שאני לא יכול לעשות. | (PERSON2) But the Flask part or the communication part with the client is something that I cannot do. | en-he |
כי אין לי ניסיון ואין לי זמן לעשות את זה. | Because I have no experience and no time to get it running. | en-he |
(דמות1) חשבתי שהכוונה היא להשתמש לא יודע, במתווך. | (PERSON1) I thought it meant something using I don't know mediator. | en-he |
(דמות2) לא, תיצור לקוח HTTP שיכול למלא את התור ולקרוא מתוך תור אחר. | (PERSON2) No, make a HTTP client which can fill the queue and read from another queue. | en-he |
ותשלח חזרה את הטקסט בפורמט של בקשת HTTP. | And send the text back in a HTTP request. | en-he |
כי הרעיון שלי הוא שהלחיצות של המשתמש ישמשו גם בפרויקט האוקראיני. | Because my idea is that the user clicks be used in the Ukrainian project too. | en-he |
(דמות2) והרעיון שלי הוא שלקוח לוחץ על הכפתור הזה של המיקרופון ומתחיל לדבר, וכשהמשתמש מדבר, אתה שולח את האודיו ביחידות קטנות בו זמנית. | (PERSON2) And my idea is that a client clicks this microphone button and starts to speak, and when the user is speaking, you send the audio in small chunks at the same time. | en-he |
נניח 200 או אפילו 100 מילישניות. | Let's say 200 or even 100 milliseconds. | en-he |
ובנוגע לשרת HTTP שים אותו בתור אחר ואני אטפל בתור הזה מאוחר יותר (ASR). | And to the HTTP server put it to a queue which I will handle the queue later with (ASR). | en-he |
והמודל יהיה קצת הגיוני, בממשק עם זה. | And the model will do some logic, some inference with that. | en-he |
ובכל פעם שיש לו משהו חדש לתמלל, הוא ישלח אותו בחזרה, והמשתמש צריך לקבל את התמליל החלקי בחזרה. | And whenever it has a new piece of transcript, it will send it back, and the user should get the partial transcript back. | en-he |
וזה צריך להיות מוצג מיד בשדה של הטקסט. | And it should be immediately displayed in the text field. | en-he |
(דמות1) יכולתי לחשוב על זה. | (PERSON1) I could think about that. | en-he |
רק שאני גם לא ממש מכיר את Flask, אבל כן, אני יכול לחשוב על זה. | Just I don't really know Flask either, but yeah, I can think about that. | en-he |
(דמות2) אה, אולי שרת HTTP אחר. | (PERSON2) Oh maybe a different HTTP server. | en-he |
אני לא יודע, אולי אנחנו אפילו לא צריכים שרת HTTP יותר מדי חזק בשביל זה. | I don't know, maybe we don't even need a fancy HTTP server for that. | en-he |
(דמות1) אני מתכוון בשביל הזרימה של הטקסט, אנחנו משתמשים במשהו כמו Hypercore, אבל אני לא בטוח אם זה יעבוד בשביל זה. | (PERSON1) I mean for text flow, we use something like Hypercore, but I'm not sure if that would work for this. | en-he |
(דמות2) כן, הדרישה היחידה היא שהוא יתמודד עם כמה חיבורים בו זמנית. | (PERSON2) Yeah, the only requirement is that it should handle more connection at a time. | en-he |
(דמות1) אני חושב שזרימת טקסט באונליין היא בערך 2,000 לפחות. | (PERSON1) Well I think online text flow is something around 2 000 at least. | en-he |
(אדם1) אני מתכוון שהסקריפט שיש לי בשני תיקיות שונות, ואז אני אפעיל את זה עם עובדים פנימיים קצת שונים, אז אני מתכוון, כן, בוא נאמר <לא מובן/> עשר, במקום פלט, קובץ של <לא מובן/> עובד צדדי, אז <לא מובן/> שיש לנו, אז אני רק אוודא שהדברים האלה עבור קובץ יחיד נוצרים במקביל מכל העובדים הזמינים. | (PERSON1) I mean the script that I have in two different folders, and then will run it with the little different inside workers, so that I mean, yeah, so that I have <unintelligible/> let's say ten, well instead of an output, a file by a <unintelligible/> side worker, so <unintelligible/> that we have, so I'll just mak... | en-he |
(אדם1) זה יחסוך לנו זמן. | (PERSON1) Oh that would save our time. | en-he |
(אדם7) כן, כן, זה מאוד חשוב למקבל את הכל בעצם, למקבל גם את העיבוד של המערכות וגם את ההערכה של התא DF, כי הוא יצטרך להריץ את זה הרבה פעמים, יהיו שגיאות, אתה צריך להריץ את זה שוב מהר. | (PERSON7) Yeah, yeah, this is very important to parallelize everything actually, parallelize both the processing of the systems as well as the evaluation of the cell DF, because he would need to run this many times, there will be errors, you need to re-run this quickly. | en-he |
(אדם7) אוקיי, אז תודה. | (PERSON7) Okay, so thanks. | en-he |
אז אתה מאוד עסוק עם זה, אני לא רוצה להעמיס עליך עוד דברים, אז בוא נתחיל עם התזכורות. | So, you are very busy with this, I don't want to put any other things on you, so let me start with the reminders. | en-he |
אנחנו צריכים לסיים את התיאור של סט השולחן של [פרויקט1], | We need to finish the description of [PROJECT1] desk set, | en-he |
מי איתנו כאן? | Who do we have here? | en-he |
אוקיי, לא באמת, כן (זה נעשה) וזה לא פה לצערי. | Oh well, not really, yeah, (it's done) and it's not here unfortunately. | en-he |
אז זה יהיה עבור [אדם2], | So, this would be for [PERSON2], | en-he |
ואז ההערכה, היא עבור [אדם1], אבל דברנו על זה. | Then the evaluation, this is for [PERSON1], but we have discussed this. | en-he |
תראה, תן לי את הטבלה ב-DSD, בוא נמתח קו עדין כאן ונקדם את התבניות של הטבלה, כמו מבנה של תכנית ל-delebl כך שהמספרים יגיעו אחר כך, אבל אני רוצה לראות קצת מספרים ואת המבנה המלא של הטבלה בקרוב. | Listen, give me the table in DSD let's put the fine line here and push the template of the table, like the plan structure into the delebl now, so that the numbers will come later, but I want to see some numbers and the full structure of the table very soon. | en-he |
(אדם7) אוקיי, תודה. | (PERSON7) Okay, thank you. | en-he |
יש לנו את [אדם5], זה טוב. | We have [PERSON5], that's good. | en-he |
אז, עבור [אדם5], אני רוצה לבדוק את ההתקדמות בהערכה של המערכות ואם אתה בקשר עם [אדם1] גם עבור האבולוציה, שהיא חשובה יותר. | So, for [PERSON5], I would like to check what is the progress in the evaluation of the systems, and whether you are in touch with [PERSON1] also for the evolution, which is more important. | en-he |
אתה יכול לשמוע אותנו, [אדם5]? | Can you hear us, [PERSON5]? | en-he |
(אדם1) ואז, נחשוב איזה חבילות להתקין בסביבת Petri. | (PERSON1) and then, we'll figure out which packages to install in the Petri environment. | en-he |
(אדם7) כן, זה עוד דבר, | (PERSON7) Yeah, so there's another thing, | en-he |
אולי זה לא רעיון טוב לשלב את שני ההרצות האלה בהגדרה אחת. | Maybe it's not a good idea to combine these two runs into one setup. | en-he |
אני חושב שזה בעצם יותר טוב להפעיל את זה בנפרד ולשמור את הפלטים. | I think it is actually better if you run it separately and save the outputs. | en-he |
בהתחלה אתה משתמש בסביבה כבקרת שיוט והעיבוד, | So, you first use the environment for cruise control and the processing, | en-he |
(אדם7) ואתה אוסף את כל קובצי הפלטים ואז מריץ את ההערכה. | (PERSON7) and you collect all the output files, and then you run the evaluation. | en-he |
והסיבה להפרדה הזאת היא שאפשר לשתק יותר. | And the reason for this split is that you can paralyze more. | en-he |
אז יהיו בעיות בכל מיני חזיתות, אני בטוח, ואנחנו נצטרך את המספרים בקרוב מאוד. | So, there will be problems on various fronts, I'm sure, and we will need the numbers very soon. | en-he |
אז, אם לדוגמה, חלק מקובצי הבדיקה ייכשלו מסיבה כלשהי, | So, if for example, some of the test files will fail for some reason, | en-he |
אז עדיין אפשר להשאיר את התרגום פעיל עבור הנותרים, | Then you can still keep the translation running for the remaining ones, | en-he |
ולהעריך מחדש, עם sel TF, את החסרים וכן הלאה. | and re-evaluate, with a sel T F, the missing ones, and, and so on. | en-he |
אז הנקודה היא לא ליצור עיבוד סדרתי, סידורי שכולל את שניהם, | So, it's, the point is not to create a serial, sequential processing that includes both, | en-he |
אלא לקבל את התוצאות וזה בעצם טוב יותר לעשות את זה בחלקים. | but the point is to get the results, and it's actually better to do it in slices. | en-he |
תחילה לקבל את כל הפלטים ואז לקבל את כל ההערכות. | First get all outputs, then we get all evaluations. | en-he |
(אדם1), כן, אני מתכוון, עכשיו, עשיתי את אותו דבר. | (PERSON1) Yeah, so, I mean, right now, I was doing the same. | en-he |
(אדם2) כן, אלה הדברים היחידים שעשיתי עבור אילוצים חיוביים. | (PERSON2) Yes, that's the only stuff I did for positive constraints. | en-he |
עם האילוצים השליליים עשיתי משהו קצת יותר מסובך בקידוד. | With the negative constraints I did something a little bit more complicated in the decoding. | en-he |
כמו לסנן החוצה את הקורות והחלפתם עם קורות שונות. | Like filtering out the beams and replacing the beams with different beams. | en-he |
ועבור אילוצים מרובי-סמלים ודברים כאלה. | And for like multi-token constraints and this kind of stuff. | en-he |
אבל עבור האילוצים החיוביים, רק השוויתי את הגישה של ההכשרה עם, כן, הדברים הפשוטים האלה. | But for the positive constraints, I only compared the training approach with yeah just this very simple stuff. | en-he |
רק שינוי של הניקוד בכל שלב באותה דרך. | Just modifying the scores in each step in the same way. | en-he |
(אדם7), אוקיי, אז בחזרה לאילוצים החיוביים. | (PERSON7) Okay, so back to the positive constraints. | en-he |
האם כבר התמקדתם בניתוח המערכת? | Were you already focusing on the analysis of the system? | en-he |
(אדם2) כן, עכשיו, אני בעצם מחכה לעיבוד מחדש של מערך הנתונים מ-([אדם9]). | (PERSON2) Yeah, right now, I'm kind of waiting for the dataset reprocessing from ([PERSON9]). | en-he |
גם [אדם9] הסתכלתי על ה(קוד) ואני חושב שאני יודע למה זה כל כך איטי. | Also [PERSON9] I had a look at the (code), and I think I know why it's so slow. | en-he |
מפני שבכל ערך בטרמינולוגיה בעיבוד של (Europar), אתה פותח את מערך הנתונים, מסמל אותו, קורא אותו שורה אחר שורה, מסמל אותו שורה אחר שורה ואז סוגר אותו. | Because for each entry in the terminology in the (Europar) reprocessing, you open the dataset, tokenize it, read it line by line, tokenize it line by line and then close it. | en-he |
אבל אתה רק עושה את זה – | But you only you do that - | en-he |
(אדם4) אני זוכר שעבור [פרויקט1] עשיתי את זה בשנה שעברה מאיזה קוטג' בהרים. | (PERSON4) So, I remember that for [PROJECT1] I was doing this from last year from some cottage in the mountains. | en-he |
ורק עכשיו הבנתי, קצת לפני ערב השנה החדשה, שאני צריך להגיש את זה. | And I just realized, shortly before New Year’s Eve that I need to submit it. | en-he |
ו-<צחוק/> אז זה גם משהו שיקרה ל-[אדם8] אני חושש שהוא יכול... לא במסיבה, אבל אולי במקום רחוק. | And <laugh/> so this is also something that will happen to [PERSON8] I'm afraid that he could be...well, no party, but maybe at the remote party. | en-he |
ואז פתאום הוא יבין שהוא צריך להגיש את זה. | And then suddenly he would realize that he has to submit it. | en-he |
אז לא תהיה בדיקה מהסוג הזה בנקודה הזאת. <צחוק/> | So, there will be no review of like at that point. <laugh/> | en-he |
אז הוא פשוט ייקח את ה-PDF ויעלה אותו. <צחוק/> | He would simply take the PDF and upload it so. <laugh/> | en-he |
(אדם7) אוקיי. | (PERSON7) Well, okay. | en-he |
אז, כן, אני אבדוק ב-21, אחרת אשלח לכם אימייל. | So, Yeah, I will check on 21st, and otherwise I will send them an e-mail. | en-he |
(אדם4) בבקשה, אני חושב שעדיך לשלוח להם אימייל עכשיו ולבקש את הקישור מראש. | (PERSON4) Please, I think better send them an e-mail right now and just ask for that link in advance. | en-he |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.