question
string
paper_id
string
chunk_index
int32
chunk
string
label
int32
answerable
bool
spans
dict
source
string
retrieval_rank
int32
gold_paper
string
gold_chunk
int32
predicted_texts
list
latency_s
float32
err
string
skipgram model character n-grams word representations
P16-2023
7
## 6 Results The results are reported in terms of perplexity, in Table 1 for English, and in Table 2 for FlemishDutch. We computed baseline perplexity scores with SRILM (Stolcke, 2002) for 1bw. We used an interpolated modified Kneser-Ney language model, with Good-Turing discounting to mimic our thresholding options. A...
1
true
{ "start": [ 1738 ], "end": [ 2616 ], "text": [ "Upon inspection of the model sizes, we observe that the skipgram model contains almost five times as many parameters as the n -gram model. This difference is explained by the addition of skipgrams of length 3 and 4, and the bigrams and unigrams deri...
retrieved
1
Q17-1010
17
[ "Upon inspection of the model sizes, we observe that the skipgram model contains almost five times as many parameters as the n -gram model. This difference is explained by the addition of skipgrams of length 3 and 4, and the bigrams and unigrams derived from these skipgrams. Each 4gram can be deconstructed into thr...
3.496642
skipgram model character n-grams word representations
2021.ranlp-1.34
10
## 4.2 FastText Based Model FastText 7 is an open-source library, developed by Facebook AI Research lab with the purpose of text classification and representation. As Bojanowski et al. (2016) described in their work, fastText creates word representations based on the skipgram model, where each word is represented as a...
1
true
{ "start": [ 165 ], "end": [ 1004 ], "text": [ "As Bojanowski et al. (2016) described in their work, fastText creates word representations based on the skipgram model, where each word is represented as a bag of character n-grams. A vector representation is associated to each character n-gram, word...
retrieved
2
Q17-1010
17
[ "As Bojanowski et al. (2016) described in their work, fastText creates word representations based on the skipgram model, where each word is represented as a bag of character n-grams. A vector representation is associated to each character n-gram, words being represented as the sum of these representations. Using ch...
3.496642
skipgram model character n-grams word representations
P16-2023
6
## 5 Experimental Setup We train 4-gram language model on the two training corpora, the Google 1 billion word benchmark and the Mediargus corpus. We do not perform any preprocessing on the data except tokenisation. The models are trained with a HPYLM. We do not use sentence beginning and end markers. The results for t...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
Q17-1010
17
[]
3.496642
What is the Revision module?
W19-4502
21
## 5.8 Module: Revision While the previous modules handle major tasks, a processed locution may still need additional adjustments, including grammar correction. Hence, the Revision module makes adjustments to a processed locution and outputs proposition(s). This task is formulated as a seq2seq problem, i.e., a model a...
1
true
{ "start": [ 25 ], "end": [ 395 ], "text": [ "While the previous modules handle major tasks, a processed locution may still need additional adjustments, including grammar correction. Hence, the Revision module makes adjustments to a processed locution and outputs proposition(s). This task is formu...
gold
-1
W19-4502
21
[ "While the previous modules handle major tasks, a processed locution may still need additional adjustments, including grammar correction. Hence, the Revision module makes adjustments to a processed locution and outputs proposition(s). This task is formulated as a seq2seq problem, i.e., a model automatically learns ...
2.198435
What is the Revision module?
2020.lrec-1.111
8
## 4.2.3 Module 3: Revision Detector It is mainly a manuscript viewer, which can be used directly after pre-processing the data. It will highlight three different kinds of revisions in manuscripts. Crossed out areas, annotations made above a text line, and probable changes of single letters (e.g. if a scribe changes t...
1
true
{ "start": [ 38 ], "end": [ 772 ], "text": [ "It is mainly a manuscript viewer, which can be used directly after pre-processing the data. It will highlight three different kinds of revisions in manuscripts. Crossed out areas, annotations made above a text line, and probable changes of single lette...
retrieved
1
W19-4502
21
[ "It is mainly a manuscript viewer, which can be used directly after pre-processing the data. It will highlight three different kinds of revisions in manuscripts. Crossed out areas, annotations made above a text line, and probable changes of single letters (e.g. if a scribe changes the letter <a> to <e>). The model ...
2.198435
What is the Revision module?
P11-4017
4
## 4 Efficient Access to Revisions Even though article revisions are available from the official Wikipedia revision dumps, accessing this information on a large scale is still a difficult task. This is due to two main problems. First, the revision dump contains all revisions as full text. This results in a massive amo...
1
true
{ "start": [ 473 ], "end": [ 555 ], "text": [ "Thus, we have developed a tool called RevisionMachine , which solves these issues." ] }
retrieved
2
W19-4502
21
[ "Thus, we have developed a tool called RevisionMachine , which solves these issues." ]
2.198435
What is the Revision module?
P11-4017
9
## Acknowledgments This work has been supported by the Volkswagen Foundation as part of the Lichtenberg-Professorship Program under grant No. I/82806, and by the Hessian research excellence program 'Landes-Offensive zur Entwicklung Wissenschaftlich-¨ okonomischer Exzellenz' ( LOEWE ) as part of the research center 'Di...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
W19-4502
21
[]
2.198435
mental health models features
2021.clpsych-1.19
5
## 4.1 Topic Model Analysis Model. We use a topic model analysis to identify topic distribution differences between demographic groups. We train on each dataset (separately) a Partially Labeled LDA model (Ramage et al., 2011), which incorporates per-label latent topics into an LDA model. We assign both depression and ...
1
true
{ "start": [ 36 ], "end": [ 679 ], "text": [ "We use a topic model analysis to identify topic distribution differences between demographic groups. We train on each dataset (separately) a Partially Labeled LDA model (Ramage et al., 2011), which incorporates per-label latent topics into an LDA model...
gold
-1
2021.clpsych-1.19
5
[ "We use a topic model analysis to identify topic distribution differences between demographic groups. We train on each dataset (separately) a Partially Labeled LDA model (Ramage et al., 2011), which incorporates per-label latent topics into an LDA model. We assign both depression and demographic labels to individua...
6.166574
mental health models features
2025.c3nlp-1.10
1
## 1 Introduction Over 197 million individuals in India are diagnosed with mental health disorders (Sagar et al., 2020), a disproportionate majority of whom do not receive mental healthcare (Singh, 2018). Generative AI technologies can facilitate affordable and easily accessible mental health assessment and support, e...
1
true
{ "start": [ 1815 ], "end": [ 2641 ], "text": [ "To inform the research on culturally competent mental health models (Sue, 1998), we adopt interpretable features that are comprehensible to stakeholders such as psychologists and policymakers, for modeling cross-cultural variations in mental health ...
retrieved
3
2021.clpsych-1.19
5
[ "To inform the research on culturally competent mental health models (Sue, 1998), we adopt interpretable features that are comprehensible to stakeholders such as psychologists and policymakers, for modeling cross-cultural variations in mental health language. We use psychosocial word categories (e.g., Linguistic In...
6.166574
VideoCLIP video text transformer properties
2021.emnlp-main.544
19
## 6 Conclusion We have presented VideoCLIP, an approach to pretrain a video-text model for zero-shot transfer to end tasks that require fine-grained association between video and language. VideoCLIP uses an objective that contrasts temporally overlapping positives with hard negatives stemming from nearest neighbor re...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
2021.emnlp-main.544
5
[]
3.256081
VideoCLIP video text transformer properties
2021.emnlp-main.544
4
## 3 VideoCLIP Pre-training In the paradigm of multi-modal video-text pretraining for zero-shot transfer, the key challenge is to learn fine-grained association in-between video and text to cover the diverse needs of end tasks. We cover VideoCLIP pre-training in this section, and discuss the needs of zero-shot transfe...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
2021.emnlp-main.544
5
[]
3.256081
12 documents annotated signal instances
W19-2708
7
## 3.2 The Signal Annotation System Prior to this work, rstWeb had no support for signal annotation. The contribution of the present work was to build a signal annotation system on top of rstWeb to allow annotators to view and edit signal annotations and make these available for export and use in downstream tasks. In...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
W19-2710
7
[]
1.585181
12 documents annotated signal instances
W19-2708
8
## 3.3 Data Model A signal in our system consists of four elements: 1. A relation whose type (RESULT, CONCESSION, etc.) the signal is helping to indicate 2. A possibly empty list of tokens which comprise the signal 3. A type that categorizes the signal according to its linguistic nature 4. A more fine-grained subtype...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
W19-2710
7
[]
1.585181
12 documents annotated signal instances
W19-2710
4
## 2.2 The Signal Anchoring Mechanism As mentioned in Section 1, RST-SC does not provide information about the location of discourse signals. Thus, Liu and Zeldes (2019) presented an annotation effort to anchor signal tokens in the text, with six categories being annotated. Their results showed that with 11 documents ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
W19-2710
7
[]
1.585181
12 documents annotated signal instances
D19-1680
6
## 4.3 Annotating intervention instances We provided definitions and text examples for the intervention types 7 to two annotators, and then asked them to identify and annotate intervention instances for each document. Annotators are provided with a User Interface (UI) (Chan et al., 2019) which allows them to search fo...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
W19-2710
7
[]
1.585181
random indexing
W11-4631
3
One important advantage of the particular random indexing approach is that the full n × d feature matrix F never needs to be explicitly computed or represented (Karlgren and Sahlgren, 2001). As described above, with RI we construct the representation of the data in G by incrementally accumulating the index vectors as...
1
true
{ "start": [ 2 ], "end": [ 590 ], "text": [ "One important advantage of the particular random indexing approach is that the full n × d feature matrix F never needs to be explicitly computed or represented (Karlgren and Sahlgren, 2001). As described above, with RI we construct the representation of...
retrieved
1
S16-2024
13
[ "One important advantage of the particular random indexing approach is that the full n × d feature matrix F never needs to be explicitly computed or represented (Karlgren and Sahlgren, 2001). As described above, with RI we construct the representation of the data in G by incrementally accumulating the index vectors...
2.674618
random indexing
W11-4631
10
## References - Dimitris Achlioptas. 2001. Database-friendly random projections. In Proceedings of the Twentieth ACM SIGMOD-SIGACT-SIGART Symposium on Principles of Database Systems , Santa Barbara, USA. - Bernd Bohnet. 2010. Top accuracy and fast dependency parsing is not a contradiction. In Proceedings of the 23rd I...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
S16-2024
13
[]
2.674618
Hindi corpus 65 phrases number of classes
C69-6214
0
Abstract AN APPLICATION OP COMPUTER TECHNIQUES TO ANALYSIS OF THE VERB PHRASE IN HINDI AND ENGLISH: A Preliminary Report Dr, LoM, Khubohandanl and WoW. Glover Authors worked on the Project at ~oona, India with the facilities of the computer CDC 3600-160A installed at the Tats Institute for Fundamental Research, Bomb...
1
true
{ "start": [ 884 ], "end": [ 1074 ], "text": [ "The results obtained with a criterion for classification of \"identical context one-deep on both sides\" were quite satisfactory. In Hindi 25 classes were formed from the corpus of 65 phrases." ] }
gold
-1
C69-6214
-1
[ "The results obtained with a criterion for classification of \"identical context one-deep on both sides\" were quite satisfactory. In Hindi 25 classes were formed from the corpus of 65 phrases." ]
0.368831
Hindi corpus 65 phrases number of classes
I11-1013
5
## 3.2 Verb Classing Using a segmenter, the root verb is separated from its inflected suffix for all the extracted verb phrases. These extracted verb phrases are then clustered based on the root verb so that all the variations of a root verb ' < verb > ' are grouped together into one cluster. As an example, a part of ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
C69-6214
-1
[]
0.368831
Hindi corpus 65 phrases number of classes
I08-5004
12
## 8.2 Results on the Test Data The best identified feature set is used for the development of the NER systems for all the five languages. We have already mentioned that for only for Bengali and Hindi we have added linguistic rules and gazetteer lists in the MaxEnt based NER systems. The accuracy of the system on the...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
C69-6214
-1
[]
0.368831
Hindi corpus 65 phrases number of classes
W17-0118
12
Table 5: Class sizes. When there is no example in the curated set (C), an example is taken from the hypothesized set output by the system. | Class | Example | C | GL (in C) | CB (in C) | |---------|--------------------|-----|-------------|-------------| | 0001 | tatuk 'have fever' | 0 | 29 (1) | 11 (0) | | 0010 | ta...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
C69-6214
-1
[]
0.368831
Hindi corpus 65 phrases number of classes
I11-1013
7
## 3.4 Generation of Verb Phrase Dictionary Given the root verb mapping and the classes to which these source and target root verb belong to, we create a 'source class' to 'target class' mapping, or a 'verb-pair class', by replacing the root verbs with their corresponding verb classes. This causes each of the verb pai...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
C69-6214
-1
[]
0.368831
research approach discussed in the paper
W94-0307
0
## Content and Rhetorical Status Selection in Instructional Texts Leila Kosseim kosseim@iro.umontreal.ca Guy Lapalme lapalme@iro.umontreal.ca D4partement d'informatique et de recherche op4rationnelle Universitd e Montrdal PB 6128, Succ. Centre Ville Montrdal, Qudbec, Canada H3C 3J7 be represented as in figure 1...
1
true
{ "start": [ 418 ], "end": [ 909 ], "text": [ "This paper discusses an approach to planning the content of instructional texts. The research is based on a corpus study of 15 French procedural texts ranging from step--bystep device manuals to general artistic procedures. The approach taken starts f...
gold
-1
W94-0307
-1
[ "This paper discusses an approach to planning the content of instructional texts. The research is based on a corpus study of 15 French procedural texts ranging from step--bystep device manuals to general artistic procedures. The approach taken starts from an AI task planner building a task representation, from whic...
3.18098
research approach discussed in the paper
O14-5000
52
## B.1 背景文步 follow NE ( CD ) , NE ( CD ) show that NE ( CD ) demonstrate that NE ( CD ) propose model it be , however , there be , however , to knowledge , there be to good of knowledge , in case , however , NE ( CD ) present NE ( CD ) describe however , in case , to knowledge , this be collection comprise CD in pract...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
W94-0307
-1
[]
3.18098
research approach discussed in the paper
2023.findings-emnlp.182
19
## Limitations In this paper, we have discussed two prominent sources of hallucination for LLMs in natural language inference tasks. We acknowledge that this is not an exhaustive search of all the sources, where further exploration should be done in future work. We also note that after controlling for the factors dis...
1
true
{ "start": [ 570 ], "end": [ 695 ], "text": [ "As discussed in Appendix A, we compared a range of popular LLM prompting techniques and selected the most promising approach." ] }
retrieved
2
W94-0307
-1
[ "As discussed in Appendix A, we compared a range of popular LLM prompting techniques and selected the most promising approach." ]
3.18098
research approach discussed in the paper
2023.nlp4dh-1.12
6
## 2.3 Research process Figure 1 presents the key phases in the research process. During the study, I conducted both distant reading and traditional close reading in parallel (Jänicke et al., 2015). During the distant reading phase, I utilised computational methods to analyse the material based on topics and keywords,...
1
true
{ "start": [ 25, 912 ], "end": [ 865, 1462 ], "text": [ "Figure 1 presents the key phases in the research process. During the study, I conducted both distant reading and traditional close reading in parallel (Jänicke et al., 2015). During the distant reading phase, I utilised computational...
retrieved
3
W94-0307
-1
[ "Figure 1 presents the key phases in the research process. During the study, I conducted both distant reading and traditional close reading in parallel (Jänicke et al., 2015). During the distant reading phase, I utilised computational methods to analyse the material based on topics and keywords, enabling a systemat...
3.18098
Transformer aspect extraction hotel reviews example
2024.stil-1.31
2
## 2. Trabalhos Relacionados Os trabalhos de identificac ¸ ˜ ao de aspectos para o portuguˆ es se baseiam, principalmente, no uso de l´ exicos [Costa and Pardo 2020], regras de linguagem [Vargas and Pardo 2020, Machado et al. 2021], em algoritmos de aprendizado de m´ aquina tradicionais [Balage Filho 2017, Vargas and ...
1
true
{ "start": [ 445 ], "end": [ 608 ], "text": [ "Em [Resplande et al. 2022], por exemplo, os autores avaliaram o uso de modelos baseados em Transformers na extrac ¸˜ ao de aspectos em avaliac ¸ ˜ oes de hot´ eis." ] }
gold
-1
2024.stil-1.31
2
[ "Em [Resplande et al. 2022], por exemplo, os autores avaliaram o uso de modelos baseados em Transformers na extrac ¸˜ ao de aspectos em avaliac ¸ ˜ oes de hot´ eis." ]
1.681122
Transformer aspect extraction hotel reviews example
2023.konvens-main.21
0
## Aspect-Based Sentiment Analysis as a Multi-Label Classification Task on the Domain of German Hotel Reviews ## Jakob Fehle Media Informatics Group University of Regensburg Regensburg, Germany jakob.fehle@ur.de ## Thomas Schmidt Media Informatics Group University of Regensburg Regensburg, Germany thomas.schmidt@ur...
1
true
{ "start": [ 715 ], "end": [ 1164 ], "text": [ "This paper addresses this gap by utilizing BERT-based transformer models, known for their exceptional performance in context-sensitive natural language processing tasks, to perform ABSA in a multi-label classification setting. We demonstrate our appr...
retrieved
1
2024.stil-1.31
2
[ "This paper addresses this gap by utilizing BERT-based transformer models, known for their exceptional performance in context-sensitive natural language processing tasks, to perform ABSA in a multi-label classification setting. We demonstrate our approach on a novel dataset of German hotel reviews that we have coll...
1.681122
Transformer aspect extraction hotel reviews example
2025.findings-acl.1273
25
## Aspect Identification: Rooms You are good at understanding documents with hotel review opinions. Below is a business review for a hotel, please extract fragments that are related to Rooms of the hotel. Definition of Rooms: Assessment of how well the room meets the guest's needs and expectations in terms of comfo...
1
true
{ "start": [ 553 ], "end": [ 632 ], "text": [ "Figure 17: The prompt of Aspect Identification for the review aspect of Rooms ." ] }
retrieved
2
2024.stil-1.31
2
[ "Figure 17: The prompt of Aspect Identification for the review aspect of Rooms ." ]
1.681122
Transformer aspect extraction hotel reviews example
2023.konvens-main.21
3
## 2 Related Work Over the last decade, ABSA has experienced significant growth through different shared task workshops, such as the SemEval Shared Tasks for the English language from 2014 to 2016 (Pontiki et al., 2014, 2015, 2016), stimulating the development of various methods addressing the three fundamental subtas...
1
true
{ "start": [ 1325 ], "end": [ 1681 ], "text": [ "In particular, since the first SemEval workshop on ABSA in 2014, the number of accessible datasets for the English language has significantly increased, covering various domains with different levels of annotation complexity, such as hotel reviews (...
retrieved
3
2024.stil-1.31
2
[ "In particular, since the first SemEval workshop on ABSA in 2014, the number of accessible datasets for the English language has significantly increased, covering various domains with different levels of annotation complexity, such as hotel reviews (Yin et al., 2017), financial microblogs (Maia et al., 2018), and A...
1.681122
What is HotpotQA
2023.acl-long.89
22
Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. 2021a. Approximate nearest neighbor negative contrastive learning for dense text retrieval. In International Conference on Learning Representations . Wenhan Xiong, Xiang Lorraine Li, Srinivasan Iyer, Jing...
1
true
{ "start": [ 575 ], "end": [ 944 ], "text": [ "Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical M...
gold
-1
2023.acl-long.89
22
[ "Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , pages 2369-2380, Brussel...
1.081433
What is HotpotQA
D18-1259
2
## 2 Data Collection The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning. One way to do so is to define reasoning chains based on a knowledge base (Welbl et al., 2018; Talmor and Berant, 2018). However, the resulting datasets are limited by the...
1
true
{ "start": [ 22 ], "end": [ 149 ], "text": [ "The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning." ] }
retrieved
1
2023.acl-long.89
22
[ "The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning." ]
1.081433
What is HotpotQA
2022.coling-1.518
7
## 4 Data Analysis ## 4.1 Dataset Statistics The final dataset consists of 1,034 high-quality data samples, in which 515 samples come from the field of Biology, 401 from the field of chemistry, 88 from the field of physics, 19 from the field of electricalengineering, 7 from the field of environmentalscience, and 4 fr...
1
true
{ "start": [ 508 ], "end": [ 796 ], "text": [ "To investigate the depth and diversity of questions in KHANQ, we classify questions based on the first two words in the question and compare them to other commonly-used question generation datasets: SQuAD 2.0 (Rajpurkar et al., 2018) and HotpotQA (Yan...
retrieved
2
2023.acl-long.89
22
[ "To investigate the depth and diversity of questions in KHANQ, we classify questions based on the first two words in the question and compare them to other commonly-used question generation datasets: SQuAD 2.0 (Rajpurkar et al., 2018) and HotpotQA (Yang et al., 2018), as shown in Table 2." ]
1.081433
What is HotpotQA
D18-1259
15
## A.3 Crowd Worker Interface Our crowd worker interface is based on ParlAI (Miller et al., 2017), an open-source project that facilitates the development of dialog systems and data collection with a dialog interface. We adapt ParlAI for collecting question answer pairs by converting the collection workflow into a sys...
1
true
{ "start": [ 795 ], "end": [ 854 ], "text": [ "Figure 5: Distribution of lengths of questions in HOTPOTQA." ] }
retrieved
3
2023.acl-long.89
22
[ "Figure 5: Distribution of lengths of questions in HOTPOTQA." ]
1.081433
What is HotpotQA
D18-1259
16
## B Further Data Analysis To further look into the diversity of the data in HOTPOTQA, we further visualized the distribution of question lengths in the dataset in Figure 5. Besides being diverse in terms of types as is show in the main text, questions also vary greatly in length, indicating different levels of comple...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2023.acl-long.89
22
[]
1.081433
What is HotpotQA
2023.acl-long.89
22
Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. 2021a. Approximate nearest neighbor negative contrastive learning for dense text retrieval. In International Conference on Learning Representations . Wenhan Xiong, Xiang Lorraine Li, Srinivasan Iyer, Jing...
1
true
{ "start": [ 575 ], "end": [ 944 ], "text": [ "Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical M...
gold
-1
2023.acl-long.89
22
[ "Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , pages 2369-2380, Brussel...
0.993376
What is HotpotQA
D18-1259
2
## 2 Data Collection The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning. One way to do so is to define reasoning chains based on a knowledge base (Welbl et al., 2018; Talmor and Berant, 2018). However, the resulting datasets are limited by the...
1
true
{ "start": [ 22 ], "end": [ 149 ], "text": [ "The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning." ] }
retrieved
1
2023.acl-long.89
22
[ "The main goal of our work is to collect a diverse and explainable question answering dataset that requires multi-hop reasoning." ]
0.993376
What is HotpotQA
2022.coling-1.518
7
## 4 Data Analysis ## 4.1 Dataset Statistics The final dataset consists of 1,034 high-quality data samples, in which 515 samples come from the field of Biology, 401 from the field of chemistry, 88 from the field of physics, 19 from the field of electricalengineering, 7 from the field of environmentalscience, and 4 fr...
1
true
{ "start": [ 508 ], "end": [ 796 ], "text": [ "To investigate the depth and diversity of questions in KHANQ, we classify questions based on the first two words in the question and compare them to other commonly-used question generation datasets: SQuAD 2.0 (Rajpurkar et al., 2018) and HotpotQA (Yan...
retrieved
2
2023.acl-long.89
22
[ "To investigate the depth and diversity of questions in KHANQ, we classify questions based on the first two words in the question and compare them to other commonly-used question generation datasets: SQuAD 2.0 (Rajpurkar et al., 2018) and HotpotQA (Yang et al., 2018), as shown in Table 2." ]
0.993376
What is HotpotQA
D18-1259
15
## A.3 Crowd Worker Interface Our crowd worker interface is based on ParlAI (Miller et al., 2017), an open-source project that facilitates the development of dialog systems and data collection with a dialog interface. We adapt ParlAI for collecting question answer pairs by converting the collection workflow into a sys...
1
true
{ "start": [ 795 ], "end": [ 854 ], "text": [ "Figure 5: Distribution of lengths of questions in HOTPOTQA." ] }
retrieved
3
2023.acl-long.89
22
[ "Figure 5: Distribution of lengths of questions in HOTPOTQA." ]
0.993376
What is HotpotQA
D18-1259
16
## B Further Data Analysis To further look into the diversity of the data in HOTPOTQA, we further visualized the distribution of question lengths in the dataset in Figure 5. Besides being diverse in terms of types as is show in the main text, questions also vary greatly in length, indicating different levels of comple...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2023.acl-long.89
22
[]
0.993376
purpose of this document
2023.wassa-1.17
11
## Acknowledgements This paper was prepared for informational purposes in part by the Artificial Intelligence Research Group of JPMorgan Chase & Co and its affiliates ('J.P. Morgan') and is not a product of the Research Department of J.P. Morgan. J.P. Morgan makes no representation and warranty whatsoever and disclaim...
1
true
{ "start": [ 21 ], "end": [ 884 ], "text": [ "This paper was prepared for informational purposes in part by the Artificial Intelligence Research Group of JPMorgan Chase & Co and its affiliates ('J.P. Morgan') and is not a product of the Research Department of J.P. Morgan. J.P. Morgan makes no repr...
gold
-1
2023.wassa-1.17
11
[ "This paper was prepared for informational purposes in part by the Artificial Intelligence Research Group of JPMorgan Chase & Co and its affiliates ('J.P. Morgan') and is not a product of the Research Department of J.P. Morgan. J.P. Morgan makes no representation and warranty whatsoever and disclaims all liability,...
1.425849
purpose of this document
J95-1002
6
## PURPOSE (taken from Mann and Thompson 1987) constraints on N: presents an activity constraints on S: presents a situation that is unrealized constraints on the N+S combination: S presents a situation to be realized through the activity in N R recognizes that the activity in N is initiated in order to realize ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2023.wassa-1.17
11
[]
1.425849
purpose of this document
2025.acl-long.1386
25
## B Detailed annotation guideline provided to the annotators ## B.1 Introduction The purpose of this work is to extract Adverse Events (AE) in elderly patients' electronic health records (discharge summaries) to be used as a novel source of data for health and social care research. In order to automate this process,...
1
true
{ "start": [ 84 ], "end": [ 770 ], "text": [ "The purpose of this work is to extract Adverse Events (AE) in elderly patients' electronic health records (discharge summaries) to be used as a novel source of data for health and social care research. In order to automate this process, we rely on Natu...
retrieved
2
2023.wassa-1.17
11
[ "The purpose of this work is to extract Adverse Events (AE) in elderly patients' electronic health records (discharge summaries) to be used as a novel source of data for health and social care research. In order to automate this process, we rely on Natural Language Processing (NLP), a sub-field of artificial intell...
1.425849
purpose of this document
2009.mtsummit-caasl.10
11
##                                 ##   ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
2023.wassa-1.17
11
[]
1.425849
purpose of this document
P79-1014
7
## Im,*|i@m|li,I@Wm~#mI~@Igm#wIiII#mmimmIII|@milIIillJgimR@ IPP does not consider the $ S H O O T script to be a total explanation of a snootin~ event. It requires a representation wnlcn indicates the purpose of the various actors, in the absence of any other information, IPP assu~es people wno shoot are deliberately ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2023.wassa-1.17
11
[]
1.425849
SimSum research project ultimate goal
W97-0714
11
## 5 Conclusion Advancing the sc~entfftc frontiers of text summanzaUon presupposes more knowledge about the way summartzatton works The mare frmt of the empmcal mvesUgat~on be- hind SlmSum is an tmage of the summanzalaon process which Is detmled enough to lay the foundattons for a stmulat~on Since the resulting summar...
1
true
{ "start": [ 133 ], "end": [ 502 ], "text": [ "The mare frmt of the empmcal mvesUgat~on be- hind SlmSum is an tmage of the summanzalaon process which Is detmled enough to lay the foundattons for a stmulat~on Since the resulting summarization model incorporates the know-how of human experts, tt has...
gold
-1
W97-0714
11
[ "The mare frmt of the empmcal mvesUgat~on be- hind SlmSum is an tmage of the summanzalaon process which Is detmled enough to lay the foundattons for a stmulat~on Since the resulting summarization model incorporates the know-how of human experts, tt has good prospects of presenting powerful techmques Summarizing by ...
1.093513
SimSum research project ultimate goal
2023.acl-long.552
21
## 9 Conclusions and Future Work In this paper, we propose SIMSUM, a new model for document-level text simplification. We demonstrate that SIMSUM sets a new state of the art on document simplification outperforming the previously competitive MUSS baseline in terms of SARI and D-SARI scores. We also release cleaned ver...
1
true
{ "start": [ 34 ], "end": [ 712 ], "text": [ "In this paper, we propose SIMSUM, a new model for document-level text simplification. We demonstrate that SIMSUM sets a new state of the art on document simplification outperforming the previously competitive MUSS baseline in terms of SARI and D-SARI s...
retrieved
1
W97-0714
11
[ "In this paper, we propose SIMSUM, a new model for document-level text simplification. We demonstrate that SIMSUM sets a new state of the art on document simplification outperforming the previously competitive MUSS baseline in terms of SARI and D-SARI scores. We also release cleaned versions of two existing large-s...
1.093513
SimSum research project ultimate goal
2023.acl-long.552
19
## 7 Human Evaluation In addition to the automatic evaluation, we performed a human evaluation of the outputs from different models. We run the assessment on 50 randomly selected samples from each dataset, thus 100 in total. We recruited two expert human evaluators to independently evaluate the generated texts from se...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
W97-0714
11
[]
1.093513
SimSum research project ultimate goal
2023.acl-long.552
39
## C /square ✓ Did you run computational experiments? 5, 6 - /square ✓ C1. Did you report the number of parameters in the models used, the total computational budget (e.g., GPU hours), and computing infrastructure used? Appendix The Responsible NLP Checklist used at ACL 2023 is adopted from NAACL 2022, with the addi...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
W97-0714
11
[]
1.093513
ELMo EEBO configuration F1 score NPR tag
2022.findings-naacl.44
16
## 10 Function tag analysis Table 6 shows the score for the function tags over the different configurations, with either 10 or 31 function tags. The overall score for the function tags drops drastically between ftags-10 and ftags31. For example, on the test sections, with ELMo EEBO it falls from a mean of 97.90% with ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2022.findings-naacl.44
12
[]
4.734863
ELMo EEBO configuration F1 score NPR tag
2025.africanlp-1.31
8
## 4.3 Evaluation We use F 1 score to evaluate our models. We only evaluate morphological tagging performance, as opposed to full morphological parsing (segmentation + tagging). However, tagging inherently depends on segmentation in our setup, since models are trained on the pre-segmented morpheme sequences. In our m...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
2022.findings-naacl.44
12
[]
4.734863
ELMo EEBO configuration F1 score NPR tag
2025.findings-naacl.258
34
## G ELO Ratings Computation For phase I, the total number of evaluations are 36 by each annotator and we consider each annotation as a single match. In Phase II, 63 additional annotations are conducted making a total of 135 matches for computing the ELO ratings. For every match the new rating : <!-- formula-not-deco...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
2022.findings-naacl.44
12
[]
4.734863
ELMo EEBO configuration F1 score NPR tag
2022.findings-naacl.44
11
## 7 Pretraining comparison experiments Table 2 presents parsing results for the dev/test sections of the 8 cross-validation splits described in Section 4. The rows are the four embedding representations described in Section 6.1 and the columns are the F1 scores (evalb bracket scores, as discussed in Section 6.3) for ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2022.findings-naacl.44
12
[]
4.734863
Translation Quality Score definition
2025.emnlp-main.1018
3
## 3 Human re-evaluation To investigate the translation quality of the FLORES+ benchmark, we manually re-evaluated translations of the four genealogically, orthographically, and geographically diverse languages with varying resource availability: Asante Twi, Japanese, Jinghpaw, and South Azerbaijani. See Table 1 for d...
1
true
{ "start": [ 3587 ], "end": [ 4100 ], "text": [ "The original work (Goyal et al., 2022) uses a metric called Translation Quality Score (TQS) to evaluate the translation quality; however, the detailed definition of this metric is not explained in their work. For this reason, we tentatively define T...
gold
-1
2025.emnlp-main.1018
3
[ "The original work (Goyal et al., 2022) uses a metric called Translation Quality Score (TQS) to evaluate the translation quality; however, the detailed definition of this metric is not explained in their work. For this reason, we tentatively define TQS as follows:\n\n<!-- formula-not-decoded -->\n\nwhere C is the n...
2.158995
Translation Quality Score definition
2013.tc-1.6
2
## 2. Defining quality The definition of translation quality has long been an issue in academic translation studies. Much of the traditional focus in (human) translation studies has been derived from literary translation practice assuming the existence of absolute quality and the availability of unlimited resources. I...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2025.emnlp-main.1018
3
[]
2.158995
Translation Quality Score definition
2013.tc-1.6
8
## 4. Scoring Because the ability to generate scores is important in many environments, MQM provides a scoring system, defined as described below. The following basic formula is used for calculating MQM quality scores in an error-count environment: <!-- formula-not-decoded --> Where: - TQ = quality score. The overa...
1
true
{ "start": [ 15 ], "end": [ 1124 ], "text": [ "Because the ability to generate scores is important in many environments, MQM provides a scoring system, defined as described below. The following basic formula is used for calculating MQM quality scores in an error-count environment:\n\n<!-- formula-...
retrieved
2
2025.emnlp-main.1018
3
[ "Because the ability to generate scores is important in many environments, MQM provides a scoring system, defined as described below. The following basic formula is used for calculating MQM quality scores in an error-count environment:\n\n<!-- formula-not-decoded -->\n\nWhere:\n\n- TQ = quality score. The overall r...
2.158995
Translation Quality Score definition
2020.eamt-1.23
2
## 2 Background How to tell whether a translation is good or bad is one of the most important and one of the most difficult questions asked in connection with translation. Best practices for evaluating HT and MT differ, and assessments of human-machine parity have largely ignored the former. ## 2.1 Evaluation of HT ...
1
true
{ "start": [ 1139 ], "end": [ 1535 ], "text": [ "The quality score of a given translation is computed as a linear combination of error counts and severity levels (i.e., weights). The error categories are defined in the quality standard; the number of errors per category and the severity of each er...
retrieved
3
2025.emnlp-main.1018
3
[ "The quality score of a given translation is computed as a linear combination of error counts and severity levels (i.e., weights). The error categories are defined in the quality standard; the number of errors per category and the severity of each error are determined by a single qualified rater. A translation is c...
2.158995
Translation Quality Score definition
W18-2005
2
## What do we mean by translation quality? ## Well, what do we mean by 'translation'? - A translation (product) is target-language content corresponds to source-language content - Must include text - May include non-textual elements such as audio-visual content and software components - Translation (process) is the a...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
2025.emnlp-main.1018
3
[]
2.158995
open access scientific publications definition
J08-4008
1
## 1. Why Open Access? There are a number of definitions of the term 'open access' in circulation, but almost all share the key principle that scientific literature should be freely available for all to read, download, copy, distribute, and use (with appropriate attribution) without restriction. At the time of writing...
1
true
{ "start": [ 24 ], "end": [ 297 ], "text": [ "There are a number of definitions of the term 'open access' in circulation, but almost all share the key principle that scientific literature should be freely available for all to read, download, copy, distribute, and use (with appropriate attribution)...
retrieved
1
P00-1021
17
[ "There are a number of definitions of the term 'open access' in circulation, but almost all share the key principle that scientific literature should be freely available for all to read, download, copy, distribute, and use (with appropriate attribution) without restriction." ]
0.445962
open access scientific publications definition
2024.sdp-1.15
5
## 2.3 Scientific Article Metadata Recent datasets have made available open access publications, including their full text and figures, such as the PubMed Open Access Subset (National Library of Medicine, 2003). Datasets of scientific publications can give us access to metadata in an easy-to-use format. For example, O...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
P00-1021
17
[]
0.445962
open access scientific publications definition
2023.nlp4dh-1.16
5
## 3.1 Open Science Most academics in any discipline would agree that Open science is a good thing allowing everyone access to research results and makes these results more transparent. However, for most humanities scholars open science in practice tends to be limited to paying open-access journal publication fees. Co...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
P00-1021
17
[]
0.445962
open access scientific publications definition
2024.sdp-1.15
2
To harness the richness found in images, we propose creating a new vision text dual encoder model to improve the performance of image retrieval tasks in scientific publications. We develop a dataset of scientific image captions based on open-access articles from PubMed Open Access Subset (National Library of Medicine...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
P00-1021
17
[]
0.445962
open access scientific publications definition
Y08-1005
6
## 4. Diversion: Open Source While scientific achievement throughout history has often provided the potential for direct financial reward, that potential is great today, and is particular significant in computational linguistics. That profit potential unfortunately leads many researchers and their institutions to cont...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
P00-1021
17
[]
0.445962
example of model error in qualitative analysis
2025.banglalp-1.8
6
## 5.1 Error Analysis A comprehensive quantitative and qualitative error analysis is conducted to provide detailed insights into the proposed model's performance. ## 5.1.1 Quantitative Analysis The last row of Table 5 shows a misclassification example. Here, the model mistakenly labels a depressive text as non\_depr...
1
true
{ "start": [ 197, 548, 981 ], "end": [ 465, 959, 1507 ], "text": [ "The last row of Table 5 shows a misclassification example. Here, the model mistakenly labels a depressive text as non\\_depressive. The sentence expresses strong criticism and frustration toward systemic issues, su...
retrieved
2
P18-1135
11
[ "The last row of Table 5 shows a misclassification example. Here, the model mistakenly labels a depressive text as non\\_depressive. The sentence expresses strong criticism and frustration toward systemic issues, such as business syndicates and government accountability", "oposed model. In the first and second te...
3.551814
example of model error in qualitative analysis
2024.naacl-long.115
12
## 5.2 Qualitative Samples QUALEVAL also allows model developers to extract prominent qualitative examples that can aid in the modeling lifecycle. Given that both in an academic and industry setting, understanding representative instances of ground truth and model-generated answers is important, QUALEVAL automates tha...
1
true
{ "start": [ 748 ], "end": [ 1332 ], "text": [ "Figure 7 shows qualitative samples from the MBPP dataset generated by the DAVINCI-3 (left and center) and DAVINCI-2 (right) models. In the first example, the ground truth program uses XOR to test for uniqueness, while the generation uses a loop to ch...
retrieved
3
P18-1135
11
[ "Figure 7 shows qualitative samples from the MBPP dataset generated by the DAVINCI-3 (left and center) and DAVINCI-2 (right) models. In the first example, the ground truth program uses XOR to test for uniqueness, while the generation uses a loop to check for uniqueness. In the second example, the ground truth progr...
3.551814
Unigram assumption about words
C12-1021
10
## 3.2 Assumptions about relations between word tokens The simplest assumption about the relation between words within an utterance is probably that there is none. For a probabilistic model, this leads to a Unigram assumption about words, i.e. that the probability of a sequence of words is simply the product of the pr...
1
true
{ "start": [ 56 ], "end": [ 392 ], "text": [ "The simplest assumption about the relation between words within an utterance is probably that there is none. For a probabilistic model, this leads to a Unigram assumption about words, i.e. that the probability of a sequence of words is simply the produ...
gold
-1
C12-1021
10
[ "The simplest assumption about the relation between words within an utterance is probably that there is none. For a probabilistic model, this leads to a Unigram assumption about words, i.e. that the probability of a sequence of words is simply the product of the probability of each individual word, irrespective of ...
1.677144
Unigram assumption about words
2021.findings-acl.326
2
## 2 The Unigram Distribution The unigram distribution is a probability distribution over the possible word forms in a language's lexicon. This probability takes the frequency of a token into account, assigning larger probabilities to word forms which are more likely to be 3 As a final contribution of our work, the c...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
C12-1021
10
[]
1.677144
Unigram assumption about words
2021.findings-acl.326
3
## 2.1 Complex Vocabularies The composition of spoken vocabularies is structured according to a host of factors. Stemming from articulatory biases, each language has a set of constraints on what sequences of speech sounds can be valid words in it; this is termed the phonotactics of a language. Languages also exhibit s...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
C12-1021
10
[]
1.677144
why use competency questions in ontology authoring
W15-0402
5
From our survey of the literature, we conclude that there are few good logic-driven approaches to entailment selection. Therefore, for our interface we plan to investigate the syntactic selection method of Denaux et al. (2012), together with a preference ordering based on linear discourse structure and the user's goa...
1
true
{ "start": [ 1040, 2036 ], "end": [ 1376, 2354 ], "text": [ "To tackle this problem, we are incorporating the technique of Competency Question-driven Ontology Authoring (CQOA) (Ren et al., 2014) into our dialogue system. This new technique takes 'Competency Questions' as requirements for o...
gold
-1
W15-0402
5
[ "To tackle this problem, we are incorporating the technique of Competency Question-driven Ontology Authoring (CQOA) (Ren et al., 2014) into our dialogue system. This new technique takes 'Competency Questions' as requirements for ontologies and uses them to automatically generate authoring tests for ensuring the qua...
0.913233
why use competency questions in ontology authoring
2021.cnl-1.11
14
## Acknowledgments This work was financially supported by Hasso Plattner Institute for Digital Engineering through the HPI Research School at UCT. Many thanks to Prof BE Antia for his help with the linguistic analysis. ## References Camila Zacch´ e de Aguiar, Ricardo de Almeida Falbo, and V´ ıtor E Silva Souza. 2019...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
W15-0402
5
[]
0.913233
why use competency questions in ontology authoring
2025.ldk-1.15
16
## References Reham Alharbi, Valentina Tamma, Floriana Grasso, and Terry R. Payne. 2023. An experiment in retrofitting competency questions for existing ontologies. ArXiv , abs/2311.05662. Reham Alharbi, Valentina Tamma, Floriana Grasso, and Terry R. Payne. 2024a. An experiment in retrofitting competency questions fo...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
2
W15-0402
5
[]
0.913233
2018 Robotic Scene Segmentation Challenge
2025.naacl-long.500
18
Gupta, Vignesh Ramanathan, Viktor Kerkez, Vincent Gonguet, Virginie Do, Vish Vogeti, Vladan Petrovic, Weiwei Chu, Wenhan Xiong, Wenyin Fu, Whitney Meers, Xavier Martinet, Xiaodong Wang, Xiaoqing Ellen Tan, Xinfeng Xie, Xuchao Jia, Xuewei Wang, Yaelle Goldschlag, Yashesh Gaur, Yasmine Babaei, Yi Wen, Yiwen Song, Yuche...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2024.lrec-main.63
20
[]
0.52398
2018 Robotic Scene Segmentation Challenge
2025.naacl-long.500
17
## References Florian Barth and Tillmann Dönicke. 2021. Participation in the konvens 2021 shared task on scene segmentation using temporal, spatial and entity feature vectors. In Shared Task on Scene Segmentation . Parishad BehnamGhader, Vaibhav Adlakha, Marius Mosbach, Dzmitry Bahdanau, Nicolas Chapados, and Siva Re...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
2024.lrec-main.63
20
[]
0.52398
2018 Robotic Scene Segmentation Challenge
2025.acl-long.695
1
## Abstract Traditional reinforcement learning-based robotic control methods are often taskspecific and fail to generalize across diverse environments or unseen objects and instructions. Visual Language Models (VLMs) demonstrate strong scene understanding and planning capabilities but lack the ability to generate acti...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
4
2024.lrec-main.63
20
[]
0.52398
2018 Robotic Scene Segmentation Challenge
2025.naacl-long.500
2
## 2 Task Description The task of automatic scene segmentation was formally defined by Zehe et al. (2021a). The goal is to segment a literary text into scenes, which are parts of the text with a consistent pattern in the four dimensions time, space, character and action. A break in these dimensions corresponds to a sc...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2024.lrec-main.63
20
[]
0.52398
ACL ethical guidelines research adherence
2025.naacl-long.580
9
## 8 From Ethical Guidelines to Actionable Insights To write an ethical statement, researchers can draw inspiration from resources such as frequently asked questions on ACL ethical consideration sections 9 , the ACM Code of Ethics, guidelines for NeurIPS impact statements (Ashurst et al., 2020), and governance overvie...
1
true
{ "start": [ 53, 1610 ], "end": [ 353, 2585 ], "text": [ "To write an ethical statement, researchers can draw inspiration from resources such as frequently asked questions on ACL ethical consideration sections 9 , the ACM Code of Ethics, guidelines for NeurIPS impact statements (Ashurst et...
retrieved
1
2024.nlp4dh-1.39
17
[ "To write an ethical statement, researchers can draw inspiration from resources such as frequently asked questions on ACL ethical consideration sections 9 , the ACM Code of Ethics, guidelines for NeurIPS impact statements (Ashurst et al., 2020), and governance overviews like the European Parliament's", "When prep...
2.107032
ACL ethical guidelines research adherence
2025.emnlp-main.1442
13
## Ethical Considerations We affirm adherence to the ACL Rolling Review (ARR) ethical guidelines, explicitly addressing potential risks and responsible research practices. This research focuses on optimizing computational efficiency in large language models (LLMs), aimed at reducing resource usage and consequently low...
1
true
{ "start": [ 27 ], "end": [ 449 ], "text": [ "We affirm adherence to the ACL Rolling Review (ARR) ethical guidelines, explicitly addressing potential risks and responsible research practices. This research focuses on optimizing computational efficiency in large language models (LLMs), aimed at red...
retrieved
2
2024.nlp4dh-1.39
17
[ "We affirm adherence to the ACL Rolling Review (ARR) ethical guidelines, explicitly addressing potential risks and responsible research practices. This research focuses on optimizing computational efficiency in large language models (LLMs), aimed at reducing resource usage and consequently lowering environmental im...
2.107032
ACL ethical guidelines research adherence
2023.findings-acl.0
42
## Ethics Panel Kar¨ en Fort, Min-Yen Kan and Yulia Tsvetkov, Luciana Benotti, Mark Dredze, Pascale Fung, Dirk Hovy, Jin-Dong Kim, Malvina Nissim Tuesday, July 11, 2023 - Room: Pier 4&5 - Time: 16:15-17:45 We present our ACL Ethics Committee's progress over the last few years. Of core interest, we will present the ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
2024.nlp4dh-1.39
17
[]
2.107032
Pearson Correlation Coefficient factual robustness faithfulness
2022.findings-emnlp.267
6
Factual Robustness Following the definition above, we measure the factual robustness of a model via its success rate of attacks in a corpus. Given a test set D and a model with parameters θ , following Equation 4, the success rate of adversarial attack on D is calculated as: <!-- formula-not-decoded --> where C s (...
1
true
{ "start": [ 1316 ], "end": [ 1840 ], "text": [ "From Mix% and Incor% reported in Table 1, we can conclude that factual robustness and faithfulness have good consistency: the more factually robust the model is (lower Mix%) the better faithfulness the generated summaries exhibit (lower Incor%). Spe...
gold
-1
2022.findings-emnlp.267
6
[ "From Mix% and Incor% reported in Table 1, we can conclude that factual robustness and faithfulness have good consistency: the more factually robust the model is (lower Mix%) the better faithfulness the generated summaries exhibit (lower Incor%). Specifically, the Pearson Correlation Coefficient and Spearman Correl...
2.501833
Pearson Correlation Coefficient factual robustness faithfulness
2022.emnlp-main.501
10
## 4.3 Correlation between robustness metrics To determine how results on different robustness metrics relate to each other, we compute their correlations. These correlations should indicate how much insight we can get into the overall impact of self-rationalization on a model's robustness by only looking at select me...
1
true
{ "start": [ 881 ], "end": [ 1741 ], "text": [ "n results on pairs of evaluation metrics. Each cell color is determined by absolute value of the correlation coefficient.\n\n<!-- image -->\n\nHYP and HANS, with Pearson coefficient 0.449. Furthermore, CAD and HANS, the manually annotated challenge s...
retrieved
1
2022.findings-emnlp.267
6
[ "n results on pairs of evaluation metrics. Each cell color is determined by absolute value of the correlation coefficient.\n\n<!-- image -->\n\nHYP and HANS, with Pearson coefficient 0.449. Furthermore, CAD and HANS, the manually annotated challenge sets, show low correlation with each other, with a Pearson coeffic...
2.501833
Pearson Correlation Coefficient factual robustness faithfulness
R09-1063
4
## 2.1 Pearson's Correlation Coefficient The Pearson's correlation coefficient [9] is: <!-- formula-not-decoded --> where x is the mean of X , y the mean of Y and s x is the standard deviation of X , s y the standard deviation of Y . The correlation coefficient measures the tendency of two variables to change in va...
1
true
{ "start": [ 238 ], "end": [ 505 ], "text": [ "The correlation coefficient measures the tendency of two variables to change in value together (i.e., to either increase or decrease). r is related with the Euclidean distance, the √ 2(1 -r ) being the Euclidean distance between the standardized versi...
retrieved
2
2022.findings-emnlp.267
6
[ "The correlation coefficient measures the tendency of two variables to change in value together (i.e., to either increase or decrease). r is related with the Euclidean distance, the √ 2(1 -r ) being the Euclidean distance between the standardized versions of X and Y ." ]
2.501833
Pearson Correlation Coefficient factual robustness faithfulness
2024.inlg-main.35
13
## 6 Results and Analyses In this section, we discuss the performance of faithfulness metrics across three distinct evaluation protocols. Subsequently, we conduct a qualitative analysis through case studies. ## 6.1 Correlation Results The correlation results are demonstrated in Table 2. The following observations ca...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2022.findings-emnlp.267
6
[]
2.501833
LagunTest reading comprehension goal
2020.readi-1.10
15
- Speer, R., Chin, J., Lin, A., Jewett, S., and Nathan, L. (2018). Luminosoinsight/wordfreq: v2.2, October. - Stanovich, K. E. (1986). Matthew Effects in Reading: Some Consequences of Individual Differences in the Acquisition of Literacy. Reading Research Quarterly , 21(4):360-407. - van Heuven, W., Mandera, P., Keul...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2020.readi-1.10
14
[]
1.221176
LagunTest reading comprehension goal
2020.readi-1.10
0
## LagunTest : A NLP Based Application to Enhance Reading Comprehension ## Itziar Gonzalez-Dios, Kepa Bengoetxea, Amaia Aguirregoitia University of the Basque Country (UPV/EHU) Rafael Moreno Pitxitxi , 2. 48013 Bilbao (Bizkaia) { itziar.gonzalezd, kepa.bengoetxea, amaia.aguirregoitia } ## Abstract The ability to re...
1
true
{ "start": [ 303 ], "end": [ 1400 ], "text": [ "The ability to read and understand written texts plays an important role in education, above all in the last years of primary education. This is especially pertinent in language immersion educational programmes, where some students have low linguisti...
retrieved
2
2020.readi-1.10
14
[ "The ability to read and understand written texts plays an important role in education, above all in the last years of primary education. This is especially pertinent in language immersion educational programmes, where some students have low linguistic competence in the languages of instruction. In this context, ad...
1.221176
LagunTest reading comprehension goal
2020.readi-1.10
8
## 4.2. WordCloud tab: overall representation of the text as a wordcloud Wordclouds are visualization tools that highlight the relative frequency of words in a text. Wordclouds are very useful to quickly identify the more frequent words in the text, since they will appear bigger and bolder. This is a way to pull out t...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
2020.readi-1.10
14
[]
1.221176
hope speech detection pre-trained models MuRIL XLM-RoBERTa ALBERT BERT
2021.ltedi-1.17
5
## 3.2 Methods Based on our previous analysis of the task description and task data, combined with the characteristics of the pre-training language model, the pretraining language model we chose in this task is the XLM-RoBERTa model. The structure of the XLM-RoBERTa pre-training language model can be seen as a combin...
1
true
{ "start": [ 16, 644 ], "end": [ 234, 1133 ], "text": [ "Based on our previous analysis of the task description and task data, combined with the characteristics of the pre-training language model, the pretraining language model we chose in this task is the XLM-RoBERTa model.", "Compare...
retrieved
1
2021.ltedi-1.0
3
[ "Based on our previous analysis of the task description and task data, combined with the characteristics of the pre-training language model, the pretraining language model we chose in this task is the XLM-RoBERTa model.", "Compared with Bert, RoBERTa deletes the task of predicting the next sentence in the pre-tra...
6.995731
hope speech detection pre-trained models MuRIL XLM-RoBERTa ALBERT BERT
2021.ltedi-1.17
3
## 2 Related Work So far, many organizations and teams have invested in the identification of negative content on social media platforms and have achieved some good results (Zampieri et al., 2020; Rangel et al., 2020; Chakravarthi et al., 2020; Malmasi and Zampieri, 2017b). However, combined with our previous analysi...
1
true
{ "start": [ 1038 ], "end": [ 1536 ], "text": [ "Chakravarthi et al. used Support vector machine (SVM), Naive Bayesian (NB), k-nearest neighbors (KNN), Decision Tree (DT), Logistic Regression (LR), and other methods on the hope speech data set to get good results (Chakravarthi, 2020). Combine thei...
retrieved
3
2021.ltedi-1.0
3
[ "Chakravarthi et al. used Support vector machine (SVM), Naive Bayesian (NB), k-nearest neighbors (KNN), Decision Tree (DT), Logistic Regression (LR), and other methods on the hope speech data set to get good results (Chakravarthi, 2020). Combine their work on the hope speech data set and the code-mixing in the data...
6.995731
global lexical context features properties
C10-2078
5
## 4.1 Global Lexical Context (glc) That the lexical context might be a good indicator for the usage of an expression is obvious when one looks at examples as in (1) and (2), which suggest that literal and non-literal usages of a specific idiom co-occur with different sets of words. Nonliteral uses of break the ice (1...
1
true
{ "start": [ 511, 1267, 1869 ], "end": [ 912, 1634, 1963 ], "text": [ "t here is the global lexical context of an expression, i.e., taking into account previous and following sentences. We are specifically looking for words which are either semantically related (in a wide sense) to...
gold
-1
C10-2078
5
[ "t here is the global lexical context of an expression, i.e., taking into account previous and following sentences. We are specifically looking for words which are either semantically related (in a wide sense) to the literal or the non-\n\n2 http://maltparser.org/index.html\n\nliteral sense of the target expression...
3.153432
global lexical context features properties
C86-1028
4
## 3. Features in lexiease As mentioned above, lexical features in a lexiease grammar are of two types: contextual and non-contextual. Contextual features specify ordering and dependency relationships among major syntactic categories ('parts of speech'), agreement and government requirements, and 'selection', semantic...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
C10-2078
5
[]
3.153432
global lexical context features properties
C18-1184
14
## 3.4 Ablation Study and Performance Analysis For ablation we analyze the impact of the features and search. We also analyze the behavior of our model based on the rarity of the entities. ## 3.4.1 Feature Analysis We study the impact of features on the local and global models. For the local model, starting with onl...
1
true
{ "start": [ 218, 744 ], "end": [ 742, 1083 ], "text": [ "We study the impact of features on the local and global models. For the local model, starting with only considering the contextual evidence as described in Section 2.2.1, we see that the performance steadily increases as we add the ...
retrieved
2
C10-2078
5
[ "We study the impact of features on the local and global models. For the local model, starting with only considering the contextual evidence as described in Section 2.2.1, we see that the performance steadily increases as we add the prior and lexical features. As shown in tables 3(a) and (b) the prior and lexical f...
3.153432
global lexical context features properties
S10-1053
4
## 2.2 Feature Extraction The system can extract a variety of features to be used in training and testing. A distinction can be made between local context features and global context features . Local context features are extracted from the immediate neighbours of the occurrence of the target word. One or more of the f...
1
true
{ "start": [ 662 ], "end": [ 1491 ], "text": [ "The global context features are made up of a bag-of-words representation of keywords that may be indicative for a given word to sense/translation mapping. The idea is that words are collected which have a certain power of discrimination for the speci...
retrieved
3
C10-2078
5
[ "The global context features are made up of a bag-of-words representation of keywords that may be indicative for a given word to sense/translation mapping. The idea is that words are collected which have a certain power of discrimination for the specific target word with a specific sense, and all such words are the...
3.153432
What is HEADY?
P13-1122
0
## HEADY : News headline abstraction through event pattern clustering Enrique Alfonseca Google Inc. ealfonseca@google.com Daniele Pighin Google Inc. biondo@google.com ## Abstract This paper presents HEADY: a novel, abstractive approach for headline generation from news collections. From a web-scale corpus of Engli...
1
true
{ "start": [ 184 ], "end": [ 642 ], "text": [ "This paper presents HEADY: a novel, abstractive approach for headline generation from news collections. From a web-scale corpus of English news, we mine syntactic patterns that a Noisy-OR model generalizes into event descriptions. At inference time, w...
gold
-1
P13-1122
-1
[ "This paper presents HEADY: a novel, abstractive approach for headline generation from news collections. From a web-scale corpus of English news, we mine syntactic patterns that a Noisy-OR model generalizes into event descriptions. At inference time, we query the model with the patterns observed in an unseen news c...
1.490477
What is HEADY?
N15-1014
7
## 4 Model description: headlines as bitmaps We model headline generation as a sequence prediction task. In this manner a news article is seen as a series of observations, where each is a possible token in the document. Furthermore, each observation can be assigned to one of two categories: inheadline, or not in-headl...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
P13-1122
-1
[]
1.490477
What is HEADY?
N15-1120
13
## 5.1.4 HEADY HEADY produces a soft-clustering from a generative model, and expects the maximum number of clusters to be provided beforehand. The model then tries to approximate this number. In our experiments, 5,496 clusters were finally generated. One weak point of HEADY, mentioned above, is that lowfrequency patte...
1
true
{ "start": [ 16 ], "end": [ 505 ], "text": [ "HEADY produces a soft-clustering from a generative model, and expects the maximum number of clusters to be provided beforehand. The model then tries to approximate this number. In our experiments, 5,496 clusters were finally generated. One weak point o...
retrieved
2
P13-1122
-1
[ "HEADY produces a soft-clustering from a generative model, and expects the maximum number of clusters to be provided beforehand. The model then tries to approximate this number. In our experiments, 5,496 clusters were finally generated. One weak point of HEADY, mentioned above, is that lowfrequency patterns do not ...
1.490477
What is HEADY?
C88-1004
9
## 4.1.HEAD Feature List [HFL]: It contains information stating which f-v-pairs are considered to be members of the set HEAD. The decision of which f-v-pairm must be HF£D members is crucial to the model. In a first approach, we will adopt a pragmatical criterion. This means that we will include as HEAD features those ...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
P13-1122
-1
[]
1.490477
What is HEADY?
2023.acl-long.475
4
## 3 Methodology ## 3.1 Notations Supposing the PLM consists of L encoder layers, and MHA has H attention heads and the corresponding weights 3 are W Q h , W K h , W V h ∈ R D × D ′ , h ∈ { 1 , 2 , · · · , H } , where D refers to the hidden representation dimensions and D ′ = D H . Let the weight set of the h -th hea...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
5
P13-1122
-1
[]
1.490477
sarcasm detection training data fields
2020.figlang-1.15
3
## 3 Dataset Description The data 1 we used for model building is taken from sarcasm detection shared task of the Sec- 1 https://competitions.codalab.org/competitions/22247 ond Workshop on Figurative Language Processing (FigLang2020). There are two types of data provided by the organizers: 1. Twitter dataset and 2. ...
1
true
{ "start": [ 336 ], "end": [ 446 ], "text": [ "Training data contains the fields 'label', 'response' and 'context' and are described as shown in the Table 1." ] }
gold
-1
2020.figlang-1.15
3
[ "Training data contains the fields 'label', 'response' and 'context' and are described as shown in the Table 1." ]
1.126537
sarcasm detection training data fields
2020.figlang-1.1
5
## 3.1.2 Twitter Training Dataset For the Twitter dataset, we have relied upon the annotations that users assign to their tweets using hashtags. The sarcastic tweets were collected using hashtags: #sarcasm and #sarcastic . As nonsarcastic utterances, we consider sentiment tweets, i.e., we adopt the methodology propose...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
1
2020.figlang-1.15
3
[]
1.126537
sarcasm detection training data fields
2020.figlang-1.15
2
## 2 Related work Sarcasm is a form of figurative language where the meaning of a sentence does not hold and the interpretation is quite contrary. A quick survey about sarcasm detection and some of the earlier approaches is compiled by Joshi et al. (2017). The problem of sarcasm detection is targeted in Table 1: Fie...
1
true
{ "start": [ 308 ], "end": [ 580 ], "text": [ "Table 1: Fields used in the training data\n\n| Field | Field Description |\n|------------------------|------------------------------------------------------------------------|\n| label response context | SARCASM or NOT SARCASM Tweet or a Reddit post O...
retrieved
2
2020.figlang-1.15
3
[ "Table 1: Fields used in the training data\n\n| Field | Field Description |\n|------------------------|------------------------------------------------------------------------|\n| label response context | SARCASM or NOT SARCASM Tweet or a Reddit post Ordered list of dialogue |" ]
1.126537
sarcasm detection training data fields
2020.clicit-1.22
4
## 4 Data Set We use the Twitter Corpus from the CodaLab shared task on sarcasm detection (Ghosh et al., 2020). The training data consists of 2,500 tweets labeled 'SARCASM' and 2,500 tweets labeled 'NON SARCASM', the balanced test data consists of an additional 1,800 labeled tweets. Ghosh et al. (2020), this is a self...
0
false
{ "start": [], "end": [], "text": [] }
retrieved
3
2020.figlang-1.15
3
[]
1.126537