Dataset Viewer
Auto-converted to Parquet Duplicate
audio_path
stringclasses
504 values
doc_id
stringclasses
504 values
src_text
stringclasses
504 values
src_text_system
stringclasses
1 value
src_lang
stringclasses
1 value
tgt_lang
stringclasses
4 values
domain
stringclasses
2 values
tgt_system
stringclasses
40 values
tgt_text
stringlengths
0
892
score
float64
0
100
audio
audioduration (s)
0.58
40.5
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_295.wav
2022.acl-long.597.sent_295
Then CEAS applies the generator to the four labeled documents independently and generates release notes for each class.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Anschließend wendet C-E-A-S den Generator unabhängig voneinander auf die vier Etikettendokumente an und erzeugt für jede Klasse Freisetzungsnotizen.
80
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_181.wav
2022.acl-long.468.sent_181
Hence, some question are better suited than others to the statutory article retrieval task, and the domain of the less suitable ones remains to be determined.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Daher sind einige Fragen besser geeignet als andere für die gesetzliche Aufgabe der Artikelsuche, und die Domäne der weniger geeigneten ist noch zu bestimmen.
95
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_298.wav
2022.acl-long.597.sent_298
We modeled the classwise abstractive summarization approach by two different methods.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Wir modellieren den grasweisen, obstruktiven Summize-Ansatz mit zwei verschiedenen Methoden.
75
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_113.wav
2022.acl-long.468.sent_113
Common natural language for the questions and complex legal language for the statutes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
die für die Fragen eine gemeinsame natürliche Sprache und eine komplexe Rechtssprache für die Statuten sind.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_71.wav
2022.acl-long.410.sent_71
But we have some extra information like seventeen fewer pictures, and Steven has eight pictures, which is totally irrelevant.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
aber wir haben einige zusätzliche Informationen siebzig fühlen Frieden und Stephen haben acht Frieden, die völlig irrelevant ist
54
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_224.wav
2022.acl-long.567.sent_224
Perhaps more relevant for this video, we will showcase our more permissive metric, the pairwise accuracy, which measures whether the image sentence alignment score is greater for the correct image text pair than for its foiled pair.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
vielleicht relevanter für dieses Video werden wir unsere primitivere Metrik die Paarwider-Genauigkeit präsentieren, die misst, ob der Bildsatzausrichtungspunkt für das richtige Bild mehr ist als für sein Foid-Paar für
69
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_262.wav
2022.acl-long.597.sent_262
Our paper solves these two problems and automatically generates high quality release notes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
diese beiden Probleme und automatisch generieren hohe coreter ist
6
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_88.wav
2022.acl-long.410.sent_88
So, we think this sentence might be misleading the model to an incorrect predictions.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Wir denken also, dass dieser Satz das Modell in eine falsche Vorhersage führen könnte.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_180.wav
2022.acl-long.468.sent_180
For example, the tenants makes two parties a week until two AM.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Zum Beispiel gibt der Mieter zwei Partys die Woche bis 2 Uhr morgens.
95
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_299.wav
2022.acl-long.597.sent_299
The first model, which we call CAS-Single, consists of a single six to six network and generates a single release note text give a concatenation of input commit messages.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Das erste Modell, das wir G. S. Singer nennen, besteht aus einem einzigen Sequenz-zu-Sequenz-Netzwerk und generiert ein einziges langes Stück, keinen Text, bei einer Verbindung von Input- und Commit-Nachrichten.
70
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_200.wav
2022.acl-long.567.sent_200
And we do these phrase alterations by focusing on six specific pieces such as existence, plurality, counting, spatial relations, actions and entity coreference, where each piece can consist of one or more instruments, in case we found more than one interesting way to create foil instances.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
und wir machen diese Satzänderungen, indem wir uns auf sechs spezifische Stücke konzentrieren z. B. Existenzpluralität, die räumliche Beziehungsaktionen zählt, und Einheitsreferenz, bei denen jedes Stück aus einem oder mehreren Instrumenten bestehen kann. falls wir mehr als einen interessanten Weg gefunden haben, um Folieninstanzen zu erstellen
56
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_396.wav
2022.acl-long.111.sent_396
And now that the language model is already finetuned over n minus one dataset using a preliminary multitask finetuning.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Beachten Sie, dass das Sprachmodell bereits über n minus einen Datensatz mithilfe einer vorläufigen Multi-Task-Feinabstimmung fein abgestimmt ist.
78
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_148.wav
2022.acl-long.468.sent_148
Meanwhile, eighteen out of thirty two codes have less than five articles mentioned as relevant to at least one question.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Unterdessen enthalten 18 von 32 Kodizes weniger als fünf Artikel, die für mindestens eine Frage relevant sind
89
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_131.wav
2022.acl-long.468.sent_131
We collected thousands of questions annotated with categories, subcategories and legal references to relevant statutes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Wir sammelten Tausende von Fragen, annotiert mit Kategorien, Unterkategorien und juristischen Verweisen auf relevante Statuten.
79
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_138.wav
2022.acl-long.468.sent_138
Let's look at some characteristic of our dataset.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Schauen wir uns einige Merkmale aller Datensätze an.
94
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_366.wav
2022.acl-long.111.sent_366
So to finetune a language model over ah this dataset will be ineffective.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Ein Sprachmodell über diesen Datensatz zu optimieren, wird also ineffektiv sein.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_115.wav
2022.acl-long.468.sent_115
Besides, statutory law is not a stack of independent articles that can be treated as a complete source of information on their own, unlike news or recipes, for example.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Neben dem gesetzlichen Gesetz handelt es sich nicht um einen Stapel von unabhängigen Artikeln, die als komplette Informationsquelle auf eigene Faust wie Nachrichten oder Rezepte behandelt werden können.
60
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_64.wav
2022.acl-long.410.sent_64
All right. So, the best approaches are often tree based model.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
In Ordnung, die besten Ansätze sind oft ein baumbasiertes Modell.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_268.wav
2022.acl-long.597.sent_268
The left side is a commit message and the right side is the release notes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
ist eine Kometenbotschaft und die rechte Seite ist die nicht Gründe
70
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_40.wav
2022.acl-long.410.sent_40
We add it to the next state to become a new quantity.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Wir fügen es der nächsten Stufe hinzu, um ein neuer Quant zu werden.
37
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_205.wav
2022.acl-long.567.sent_205
For example, though it's not impossible, it is statistically less likely for plants to cut a man than a man to cut plants, and large vision and language models could pick up on this.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Zum Beispiel, obwohl es nicht unmöglich ist, ist es statistisch weniger wahrscheinlich, dass Pflanzen einen Menschen schneiden, als dass ein Mensch Pflanzen schneidet, und große Vision- und Sprachmodelle könnten das erkennen.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_271.wav
2022.acl-long.597.sent_271
This can be regarded as a summarization task.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Dies kann als Samanisierungsaufgabe betrachtet werden hoffentlich
45
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_77.wav
2022.acl-long.410.sent_77
So better language model has better language understanding abilities so that the number here is higher for Roberta and lower for BERT.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Ein besseres Sprachmodell hat also eine bessere Sprachverständnisfähigkeit, so dass die Zahl hier für Roberta höher und für Vögel niedriger ist.
75
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_308.wav
2022.acl-long.597.sent_308
Finally, we also calculate the specificity because ROUGE and BLEU cannot be calculated if the release notes are empty.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Schließlich korrelieren wir auch eine Spezifität. Da lus und blau nicht korreliert werden können, wenn die Rees-Notizen leer sind,
44
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_291.wav
2022.acl-long.597.sent_291
The classwise extractive then abstractive summarization model consists of two neural modules.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
extractive obstruktive einige resition moder besteht aus zwei neueren modellen ein
30
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_73.wav
2022.acl-long.410.sent_73
And we observe these two expressions actually have similar scores.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Und wir beobachten, dass diese beiden Ausdrücke tatsächlich ähnliche Werte haben.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_225.wav
2022.acl-long.567.sent_225
For more metrics and results on them, do check out our paper.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Für mehr Metriken und Ergebnisse dazu. Lesen Sie unsere Zeitung.
95
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_33.wav
2022.acl-long.410.sent_33
So this makes the process more accurate.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
dass dies den Prozess genauer macht
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_223.wav
2022.acl-long.567.sent_223
Two of our most important evaluation metrics are the accuracy of the models in classifying image sentence pairs into captions and foils.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Zwei unserer wichtigsten Bewertungskriterien sind die Genauigkeit der Modelle bei der Klassifizierung von Bild-Satz-Paaren in Bildunterschriften und FOILs.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_288.wav
2022.acl-long.597.sent_288
Also, the number of unique tokens is quite large at eight thousand eight hundred thirty thousand.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
die Anzahl der einzigartigen Token ist ziemlich groß bei achttausend achthundertdreißigtausend
50
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_307.wav
2022.acl-long.597.sent_307
This penalty results in a lower BLEU value in the experiment results described next.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Diese Buße führt zu einem unteren Bezirk. In den Ergebnissen des Experiments, die später beschrieben werden.
77
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_316.wav
2022.acl-long.597.sent_316
High coverage of CEAS can be achieved probably because the classifier can focus on selecting relevant commit messages for each class.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
hohe Abdeckung von Autos kann wahrscheinlich erreicht werden, weil der Crass-Fehler sich auf die Scting-Ereignis-Commit-Nachricht für jedes Gras konzentrieren kann viel
22
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_201.wav
2022.acl-long.567.sent_201
For example, in the case of the actions piece, we have two instruments, one in which the action verb is changed with a different action, and one in which actants are swapped.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Zum Beispiel haben wir im Fall des Aktionsstücks zwei Instrumente, eines, bei dem das Handlungsverb durch eine andere Handlung ersetzt wird, und eines, bei dem Akzente ausgetauscht werden.
70
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_307.wav
2022.acl-long.597.sent_307
This penalty results in a lower BLEU value in the experiment results described next.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
dies panay litt und über bio in den Ergebnissen des Experiments beschrieben
17
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_196.wav
2022.acl-long.567.sent_196
We target existence, plurality, counting, spatial relations, actions and entity coreference.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Existenz Pluralität Zählen von räumlichen Beziehungen Aktionen und Entity Cor Reference
10
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_255.wav
2022.acl-long.597.sent_255
The second is Glyph, recently announced in twenty twenty.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Die zweite ist Trauer. Dieser Eintrag wurde 2020 angekündigt.
32
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_69.wav
2022.acl-long.410.sent_69
So, in our prediction we find some of the intermediate values are actually negatives.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
In unserem Schutz stellen wir also fest, dass einige der Zwischenwerte tatsächlich negativ sind.
21
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_35.wav
2022.acl-long.410.sent_35
So, the expression is represented by e i j o p.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Der Ausdruck wird also durch eij op repräsentiert, wo wir Operatoren
50
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_104.wav
2022.acl-long.468.sent_104
But the majority of citizens have little to know knowledge about their rights and fundamental legal processes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Die Mehrheit der Bürger hat jedoch wenig oder gar keine Kenntnisse über ihre Rechte und grundlegenden rechtlichen Verfahren.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_101.wav
2022.acl-long.468.sent_101
Hi, my name is Antoine and I'm from Maastricht University.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
kommen danke mein name ist anton und ich m von mastick university ich
67
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_330.wav
2022.acl-long.597.sent_330
Please check out our dataset on GitHub.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Bitte schauen Sie sich unser J-Set auf GitHub an.
64
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_136.wav
2022.acl-long.468.sent_136
And each articles comes with a concatenation of the subsequence heading in the structure of the law.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
und jeder Artikel enthält eine Verkettung ihrer nachfolgenden Überschriften in der Rechtsstruktur.
81
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_87.wav
2022.acl-long.410.sent_87
So, we can actually correlate this expression with the sentence here. Alright.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Also können wir diesen Ausdruck mit dem Satz hier korrelieren.
82
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_219.wav
2022.acl-long.567.sent_219
Finetuning would only enable models to exploit artifacts or statistical biases in the data.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Eine Feinabstimmung würde es Modellen nur ermöglichen, Artefakte oder statistische Verzerrungen in den Daten auszunutzen.
91
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_360.wav
2022.acl-long.111.sent_360
So a naive approach will be ah target task finetuning.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
ein naiver Ansatz wird ein tang-task fine tuning sein so
70
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_296.wav
2022.acl-long.597.sent_296
In this task, the direct correspondences between commit messages and release notes are not known.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
In dieser Dämmerung sind die direkten Korrespondenzen zwischen den Comit-Nachrichten und diesen Notizen nicht bekannt.
78
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_356.wav
2022.acl-long.111.sent_356
But if the dataset has five classes, FeSTE generates five new features.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Wenn der Datensatz jedoch fünf Klassen hat, erzeugt er zuerst fünf neue Features,
90
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_411.wav
2022.acl-long.111.sent_411
For summing, FeSTE enables few shot enrichment from thirty five samples in our experiments.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Zusammenfassend lässt sich sagen, dass FAST eine kurzzeitige Anreicherung von 35 Proben in unserem Experiment ermöglicht.
65
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_51.wav
2022.acl-long.410.sent_51
So in the second step, we do the same thing, but the only difference is that we ah the only difference is one more quantities.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Im zweiten Schritt machen wir dasselbe, aber der einzige Unterschied ist, dass es eine Menge mehr ist.
95
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_329.wav
2022.acl-long.597.sent_329
Our experiments show that the proposed method generates less noisy release notes at higher coverage than the baselines.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Unser Experiment zeigt, daß die vorgeschlagene Methode bei einer höheren Abdeckung weniger laute Freisetzungsnoten erzeugt als die Basislinien.
90
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_263.wav
2022.acl-long.597.sent_263
With a limited applicability problem, we propose a high quality classwise summarization method using only commit messages as input.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Für das Problem der begrenzten Anwendbarkeit schlagen wir eine qualitativ hochwertige klassifizierte Zusammenfassungsmethode vor, die nur die Mitteilung des Ausschusses als Input verwendet.
80
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_128.wav
2022.acl-long.468.sent_128
Then we gathered legal questions with references to relevant statutes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Dann sammelten wir rechtliche Fragen mit Referenzen auf einschlägige Statuten.
95
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_286.wav
2022.acl-long.597.sent_286
In the end, seven thousand two hundred repositories and eighty two thousand pieces of data were collected.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Am Ende wurden 7200 gebaute Datenspeicher und 82000 Daten gesammelt.
85
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_336.wav
2022.acl-long.111.sent_336
Feature generation using another data source may add substantial information.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
die Merkmalgenerierung (unter Verwendung einer anderen Datenquelle) kann wesentliche Informationen hinzufügen.
91
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_147.wav
2022.acl-long.468.sent_147
And around eighty percent of these cited articles come from either the civil code, judicial codes, criminal investigation codes or penal codes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Und etwa 80 Prozent dieser zitierten Artikel stammen entweder aus dem Zivilgesetzbuch, dem Justizgesetzbuch, dem Strafgesetzbuch oder dem Strafgesetzbuch.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_194.wav
2022.acl-long.567.sent_194
Or do they focus on biases as shown by previous work?
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
konzentrieren sie sich auf Vorurteile, wie in früheren Arbeiten gezeigt wurde?
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_388.wav
2022.acl-long.111.sent_388
So the label vector in this case stays always ah which consists always with two classes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Also der Etikett Vektor, in diesem Fall, bleibt immer, die besteht immer, mit zwei Klassen.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_56.wav
2022.acl-long.410.sent_56
So the training procedure is similar to training a sequence to sequence model where we optimize the loss at each time step.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Das Trainingsverfahren ähnelt dem Training eines Sequenz-zu-Sequenz-Modells, bei dem wir den Verlust bei jedem Zeitschritt optimieren.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_339.wav
2022.acl-long.111.sent_339
We need an automatic process which involves entity linking and text analysis to extract new features from the knowledge base's free text.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
, benötigen wir einen automatisierten Prozess, der Verknüpfung und Textanalyse beinhaltet, um neue Funktionen aus der Wissensdatenbank zu extrahieren.
90
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_322.wav
2022.acl-long.597.sent_322
The reason for this model's reluctance is that in training data, only thirty three percent of the sentences are present in the features label and forty percent in the improvements label.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Der Grund für diese Modell-Zögerlichkeit ist, dass in Trainingsdaten nur 33% der Sätze in den Trümmern des Features und 40% in den Trümmern der Verbesserung vorhanden sind.
75
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_96.wav
2022.acl-long.410.sent_96
And the last thing is that the underlying mechanism does not only apply to network problem solving tasks but also other tasks that involve multi step reasoning.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
die letzte Sache ist, dass der zugrunde liegende Mechanismus, der nicht nur auf Mitgliedsproblemlösungsaufgaben anwendbar ist, sondern auch die Aufgabe, die mehrstufige Argumentation Wir haben
96
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_276.wav
2022.acl-long.597.sent_276
But the labels are not always consistent with each repository.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Aber die Larven stimmen nicht immer mit jedem Larvensamen überein.
36
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_81.wav
2022.acl-long.410.sent_81
And here we also show the overall performance.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Und hier zeigen wir auch die Gesamtleistung.
97
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_212.wav
2022.acl-long.567.sent_212
If an NLI model predicts the foil to contradict or to be neutral with respect to the caption, we take this as an indicator of a valid foil.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Wenn ein NLI-Modell vorhersagt, dass der FOIL der Bildunterschrift widerspricht oder neutral ist, nehmen wir dies als Indikator für einen gültigen FOIL an.
73
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_393.wav
2022.acl-long.111.sent_393
It ah it extracts the text from the knowledge base, which in this example is the abstract of the Wikipedia page.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Es extrahiert den Text aus der Wissensdatenbank, die in diesem Beispiel die Zusammenfassung der Wikipedia-Seite ist.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_392.wav
2022.acl-long.111.sent_392
And then ah FeSTE executes entity linking phase.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
dann schnell ausgeführt in die Verbindungsphase.
0
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_121.wav
2022.acl-long.468.sent_121
In this work, we present a new French native citizen-centric dataset to study whether retrieval models can approximate the efficiency and reliability of a legal expert for the task of statutory article retrieval.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
In dieser Arbeit präsentieren wir einen neuen französischen Bürgerzentrierten Datensatz, um zu untersuchen, ob das Abrufmodell die Effizienz und Zuverlässigkeit von Rechtsexperten für die Aufgabe
93
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_57.wav
2022.acl-long.410.sent_57
And here we also use this tau to represent when we should terminate this generation process.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
hier verwenden wir auch diesen Zug, um zu repräsentieren, wann wir diesen Generationsprozess beenden sollten und
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_301.wav
2022.acl-long.597.sent_301
The second method, method, which we call CAS-Multi, consists of four different seq2seq networks, each of which correspond to one of the fixed release note classes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Die zweite Methode, die wir C-E-A-S-mult nennen, besteht aus vier verschiedenen 6 - zu-6 - Netzwerken, die jeweils einer der Release-Note-Klassen entsprechen.
0
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_157.wav
2022.acl-long.468.sent_157
These embeddings typically result from a pooling operation on the output of a word embedding model.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Diese Einbettungen ergeben sich in der Regel aus einer Aggregationsoperation auf der Ausgabe eines Wort-Einbettungsmodells.
63
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_401.wav
2022.acl-long.111.sent_401
We also, we also split each dataset into four folds and apply four folds cross validation.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Außerdem wird jeder einzelne Datensatz in oder fällt aufteilen und Füße über die Validierung anwenden.
60
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_402.wav
2022.acl-long.111.sent_402
Then, we generate the new features and evaluate them using five evaluation classifiers.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Dann generieren wir die neuen Features und bewerten sie mithilfe von fünf Bewertungsklassifikatoren.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_374.wav
2022.acl-long.111.sent_374
In MTDNN, MTDNN maintains ah heads in the number of tasks in the training set.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
In leerer DNA, und leere DNA hält Köpfe bei der Anzahl der Aufgaben im Trainingssatz.
11
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_327.wav
2022.acl-long.597.sent_327
We have built a new dataset for automatic release note generation.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Wir haben ein neues J-Set für die automatische Generierung von Freisetzungsnotizen gebaut.
66
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_56.wav
2022.acl-long.410.sent_56
So the training procedure is similar to training a sequence to sequence model where we optimize the loss at each time step.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Das Trainingsverfahren ähnelt dem Training eines Sequenz-zu-Sequenz-Modells, bei dem wir den Verlust bei jedem Zeitschritt optimieren.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_141.wav
2022.acl-long.468.sent_141
The lengthiest one being up to five thousand seven hundred and ninety words.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
längste bis zu 5790 Wörter beträgt.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_215.wav
2022.acl-long.567.sent_215
Therefore, as a third measure for generating valid foils, we employ human annotators to validate the data used in VALSE.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
als dritte Maßnahme für die Erzeugung gültiger Folien verwenden wir menschliche Annotatoren, um die verwendeten Daten zu validieren beinhaltet
75
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_394.wav
2022.acl-long.111.sent_394
Then it reformulated the task into a pairwise sentence classification task.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Dann reformulierte es die Aufgabe in eine Pille senden piaclassifizierung Aufgabe
30
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_194.wav
2022.acl-long.567.sent_194
Or do they focus on biases as shown by previous work?
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
konzentrieren sie sich auf Vorurteile, wie in früheren Arbeiten gezeigt um mehr
90
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_105.wav
2022.acl-long.468.sent_105
As a result, many vulnerable citizens who cannot afford the costly assistance of a legal expert are left unprotected or, worst, exploited.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Infolgedessen sind viele gefährdete Bürger, die sich die kostspielige Hilfe eines Rechtsanwalts nicht leisten können, ungeschützt oder, schlimmer noch, ausgebeutet.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_207.wav
2022.acl-long.567.sent_207
First, we make use of strong language models to propose foils.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
wir starke Sprachmodelle, um Folien vorzuschlagen
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_333.wav
2022.acl-long.111.sent_333
And I will present our paper, Few-Shot Tabular Data Enrichment Using Fine-Tuned Transformers Architectures.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
ich werde unser Papier einige kurze Registerkarten präsentieren, die mit Intune-Transformatoren-Architektur bereichert sind Analysiert
79
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_347.wav
2022.acl-long.111.sent_347
And and the text of the entities of the knowledge base is extracted and added to the dataset.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
und der Text der Entitäten der Wissensdatenbank wird extrahiert und dem Datensatz hinzugefügt
80
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_413.wav
2022.acl-long.111.sent_413
And it keeps the head of ah of the model.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
und es hält den Kopf des Modells aber
75
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_281.wav
2022.acl-long.597.sent_281
Commit messages are not tied to each release.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Commit-Nachrichten sind nicht an jede Veröffentlichung gebunden, wie im Bild unten gezeigt.
58
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_68.wav
2022.acl-long.410.sent_68
And this dataset is challenging because the author tried to manually ah adding something to confuse the NLP model like such as adding irrelevant information and extra quantities.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
Zustand sagte, ist herausfordernd, weil der Autor versuchte, manuell etwas hinzuzufügen, um das nabi-Modell zu verwirren, wie das Hinzufügen von evaluument-Informationen und zusätzlichen Mengen
84
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_355.wav
2022.acl-long.111.sent_355
So, FeSTE generates two new features.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
so schnell neue Funktionen generieren
1
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_133.wav
2022.acl-long.468.sent_133
The remaining references were matched and converted to the corresponding article ids from our corpus.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Die verbleibenden Referenzen wurden verglichen und in den entsprechenden Artikel umgewandelt. Ideen aus „Orcopus“
81
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_17.wav
2022.acl-long.410.sent_17
So, traditional sequence to sequence model convert the expression to a specific sequence for generation.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
zu versuchen traditionelle Sequenz zu Sequenz mehr, um den Ausdruck in eine bestimmte Sequenz für die Generation zu konvertieren
50
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_23.wav
2022.acl-long.410.sent_23
So here the good thing is that it actually gives us this binary tree structure, and it is um but actually it is quite counterintuitive because we generate the operator first and then at the end we generate the quantities.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Das Gute daran ist, dass es uns tatsächlich diese binäre, wahre Struktur gibt. Und das ist es, aber eigentlich widerspricht es der Intuition, denn wir generieren zuerst den Operator und dann generieren wir am Ende die Mengen. Und die
46
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_385.wav
2022.acl-long.111.sent_385
Here is the our input dataset which consists of entities, features, text and classes.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
ist unser Inputataset, das aus Intoties-Funktionen Texte und Klassen besteht und
50
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_162.wav
2022.acl-long.468.sent_162
Siamese, which uses a unique word embedding model that maps the query and article together in a shared dense vector space, and two-tower, which uses two independent word embedding models that encode the query and article separately into different embedding spaces.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
Siamese, das ein einzigartiges Wort-Einbettungsmodell verwendet, das Anfrage und Artikel zusammen in einem gemeinsamen, dichten Vektorraum kartografiert. Und Two Towers, das zwei unabhängige Wort-Einbettungsmodelle verwendet, die den Querian-Artikel separat in verschiedene Einbettungsräume kodieren.
55
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_5.wav
2022.acl-long.410.sent_5
So this figure is taken from the PaLM paper where they perform prompting to solve the network problem in the few shot learning scenario.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
so dass diese Zahl aus dem Pel-Papier genommen wird, wo sie auffordern, das Metallproblem in wenigen kurzen Lernvorgängen zu lösen.
64
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_395.wav
2022.acl-long.111.sent_395
Applied the language model to the new task and the output likelihood for each class.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
, Anwendung des Sprachmodells auf die neue Aufgabe und die Ausgabewahrscheinlichkeit für jede Klasse.
91
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_154.wav
2022.acl-long.468.sent_154
The main problem with these approaches is that they can only retrieve articles that contain keywords present in the query.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Das Hauptproblem bei diesen Ansätzen besteht darin, dass sie nur Artikel abrufen können, die Schlüsselwörter enthalten, die in der Abfrage vorhanden sind.
93
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_309.wav
2022.acl-long.597.sent_309
A higher specificity means that the model correctly outputs an empty text in cases where the release notes assume empty.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
bedeutet eine hohe Spezifität, dass die Modelläußerungen leere Taktiken sind. In Fällen, in denen die Noten der Rees leer sind.
78
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_32.wav
2022.acl-long.410.sent_32
So, here we actually generate the whole expression directly rather than generating a single operators or quantities.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Hier erzeugen wir also den ganzen Ausdruck direkt, anstatt einzelne Operatoren oder Größen zu erzeugen.
100
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_25.wav
2022.acl-long.410.sent_25
So here if we look at this expression, eight times three plus three is actually generated twice, but in fact we should reuse the results.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=2
Also hier, wenn wir uns diesen Ausdruck ansehen, 8 mal 3 plus 3, wird er tatsächlich zweimal erzeugt. Aber in der Tat sollten wir die Ergebnisse wiederverwenden.
60
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_371.wav
2022.acl-long.111.sent_371
When you finetune the language model over the n minus one datasets.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=21
mehrere Aufgaben, Wenn wir das Sprachmodell über n Gramm abgestimmt haben, gibt es einen Datensatz.
20
data/acl6060/2/acl_6060/eval/segmented_wavs/gold/sent_232.wav
2022.acl-long.567.sent_232
From the coreference piece, we find out that tracing multiple references to the same object in an image by using pronouns is also difficult for vision and language models.
human
en
de
iwslt23:ACL
task=multilingual.year=2023.sys_id=22
dem Referenzstück erfahren wir, dass das Verfolgen mehrerer Verweise auf dasselbe Objekt in einem Bild mithilfe von Pronomen auch für Vision- und Sprachmodelle schwierig ist.
10
End of preview. Expand in Data Studio

Metrics-Shared Task IWSLT 2026 Train and Dev Set

This dataset contains the train and dev sets for the Speech Translation Metrics Shared Task at IWSLT 2026. More details about the shared task can be found on the IWSLT website.

The dataset is primarily designed for research in speech translation quality estimation.

Task Goal

Given a speech sample and a system-generated translation, the goal is to estimate a score that reflects the translation quality.


Dataset Splits

train

Contains a mix of:

  • IWSLT 2023 human annotations (details IWSLT 2023)
    • Previous and following segments can be inferred from the doc_id feature.
    • Human evaluators considered context of one previous and one following segment.
  • WMT 2024 human annotations (details WMT 2024)
    • Evaluated on segmented audio. The information on previous/following segments is not available.
  • WMT 2025 human annotations (details WMT 2025)
    • Evaluated on segmented audio. The information on previous/following segments is not available.

train_synthetic

Contains:

  • SpeechQE data (details SpeechQE)
    • Based on Common Voice.
    • Automatically annotated (synthetic scores).

dev

Contains:

  • IWSLT 2025 ACL Talks human annotations (details IWSLT 2025)
    • Previous and following segments can be inferred from the doc_id feature.
    • Human evaluators considered context of one previous and one following segment.

Features

Column Type Description
audio Audio The speech waveform of the segment
audio_path string Path to the audio file
doc_id string Unique identifier for the segment/document
src_text string Source text
src_text_system string Source text system (e.g. human, ASR model)
src_lang string Source language code (e.g., en)
tgt_text string Target text translation
tgt_lang string Target language code (e.g., de)
domain string Domain or dataset source
tgt_system string Target system or model used for translation
score float64 Human or synthetic evaluation score (0–1)

Dataset Card Contact

Maike Züfle @maikezu

Downloads last month
10