Gated Neural Network for Sentence Compression Using Linguistic Knowledge

Zhao, Yang; Senuma, Hajime; Shen, Xiaoyu; Aizawa, Akiko

doi:10.1007/978-3-319-59569-6_56

Cited by 10 publications

(11 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are mainly two different methods to solve a compression task of a sentence. While abstractive approaches [13][14] rely on paraphrasing words, extractive methods [6], [8][9], [15][16][17][18][19][20] solve sentence compression as a sequence of word deletions of the original sentence. In this case, for each word of a sentence, the compression algorithm needs to decide whether to keep or delete the word based on its given features [17].…”

Section: Related Workmentioning

confidence: 99%

“…In this case, for each word of a sentence, the compression algorithm needs to decide whether to keep or delete the word based on its given features [17]. In deletion-based sentence compression, one can distinguish between two different lines of research: one relying on manually modeled linguistic knowledge [15] and the other based on machine learning (ML) [18].…”

Section: Related Workmentioning

confidence: 99%

“…Wang, et al [17] use syntactical features in addition to word embeddings. Lai, et al [16] propose a bi-directional encoder-decoder approach while Thao, et al [18] use a gated neural network. ML models dedicated to classifying the word tokens of a sentence that will be omitted are also investigated in [19].…”

Section: Related Workmentioning

confidence: 99%

“…After these steps of preprocessing the words are finally transformed into embedded feature vectors that can be processed by neural networks. Therefore, a standard skipgram model is employed in order to transform each word of the vocabulary into a ddimensional vector [18]. Furthermore, the categorical features for a token (POS tags and DEP labels) are encoded using one-hot-encoding.…”

Section: Conceptual Designmentioning

confidence: 99%

See 3 more Smart Citations

Using Artificial Neural Networks to Derive Process Model Activity Labels from Process Descriptions

Pyrtek¹,

Hake²,

Loos³

2020

WI2020 Zentrale Tracks

View full text Add to dashboard Cite

Recently, Artificial Neural Networks (ANN) have shown high potential in the area of Natural Language Processing (NLP). In the area of sentence compression, the application of ANNs has proven to outperform existing rule-based approaches. Nevertheless, these approaches require a decent amount of training data to achieve high accuracy. In this work, we aim at employing ANNs to derive process model labels from process descriptions. Since the amount of publicly available pairs of text and process model is scarce, we employ a transfer learning approach. While training the compression model on a large corpus consisting of sentence-compression pairs, we transfer the model to the problem of deriving label descriptions. We implement our approach and conduct an experimental evaluation using pairs of process descriptions and models. We found that our transfer learning model keeps high recall while losing performance on precision and compression rate.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Conceptual Designmentioning

confidence: 99%

See 2 more Smart Citations

Using Artificial Neural Networks to Derive Process Model Activity Labels from Process Descriptions

Pyrtek¹,

Hake²,

Loos³

2020

WI2020 Zentrale Tracks

View full text Add to dashboard Cite

show abstract

“…Consequently, there is virtually nothing that provides a basis for an explanation, only input and output data being on a level accessible to humans. Some more options are available for networks with a specific topology, such as gated networks (Zhao et al 2017), where activations at intermediate levels can be visualized; but this techniques is probably suitable for a specific set of tasks only. What is remaining would be reruns with similar related data, to find out essential differences on some experimental basis.…”

Section: Neural Networkmentioning

confidence: 99%

Requirements for Conceptual Representations of Explanations and How Reasoning Systems Can Serve Them

Horacek¹

2017

Proceedings of the 1st Workshop on Explainable Computational Intelligence (XCI 2017)

View full text Add to dashboard Cite

Explanations of solutions produced by reasoning systems in ever growing complexity become increasingly interesting, which is particularly challenging in view of fundamental differences between human and machine representation and problem-solving methods. In this paper, we formulate requirements for conceptual representations that are adequate for producing human-oriented explanations, and we discuss how some reasoning mechanisms can serve them or can possibly be adapted to do so. This examination is intended to state in what ways reasoning systems can potentially support explanation generation, and where technology-justified limitations have to be accepted.

show abstract

Improving a Syntactic Graph Convolution Network for Sentence Compression

Wang

Chen

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Gated Neural Network for Sentence Compression Using Linguistic Knowledge

Cited by 10 publications

References 14 publications

Using Artificial Neural Networks to Derive Process Model Activity Labels from Process Descriptions

Using Artificial Neural Networks to Derive Process Model Activity Labels from Process Descriptions

Requirements for Conceptual Representations of Explanations and How Reasoning Systems Can Serve Them

Improving a Syntactic Graph Convolution Network for Sentence Compression

Contact Info

Product

Resources

About