PictoBERT: Transformers for next pictogram prediction

Pereira, Jayr; Macêdo, David; Zanchettin, Cleber; Oliveira, Adriano L. I.; Fidalgo, Robson

doi:10.1016/j.eswa.2022.117231

Cited by 7 publications

(12 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The literature suggests that neural networks based language models may perform better than statistical models [Goldberg and Hirst 2017]. Besides, [Pereira et al 2022] compared their model with knowledge-based approaches and demonstrated improvements. Their models outperformed the semantic grammar on predicting the correct pictogram to complete a sentence.…”

Section: Resultsmentioning

confidence: 99%

“…In the process, the codes representing the categories may be merged or renamed [Petersen et al 2015]. According to [Petersen et al 2015], the process may only be applied to the [Pereira et al 2022] Expert Systems with Applications papers' abstract. However, if the abstracts are unclear, the method may consider the paper introduction, conclusion, or other parts.…”

Section: Data Extractionmentioning

confidence: 99%

“…The authors used these vectors as input to their recurrent neural network. Other studies followed an approach similar to [Schwab et al 2020], which consider that a pictogram is better represented by a concept from a dictionary (e.g., person: a human being) [Martínez-Santiago et al 2015, Pereira et al 2020, Pereira et al 2022. This approach assumes that the concept is a link between the pictogram label and its figure.…”

Section: Prediction Unitmentioning

confidence: 99%

“…This approach assumes that the concept is a link between the pictogram label and its figure. [Martínez-Santiago et al 2015] used concepts from the FrameNet database, [Pereira et al 2020] used WordNet synsets (a set of synonyms with a glossary definition, e.g., a person is a human being), and [Pereira et al 2022] used WordNet word-senses (a link between a word and a synset). For more details about the differences between a synset and a word sense, refer to WordNet documentation 3 .…”

Section: Prediction Unitmentioning

confidence: 99%

“…For more details about the differences between a synset and a word sense, refer to WordNet documentation 3 . [Pereira et al 2022] encodes each word-sense to a real-valued vector using the embeddings constructed by [Scarlini et al 2020]. Approaches based on concepts (synsets, word-senses) may fit better polysemic words.…”

Section: Prediction Unitmentioning

confidence: 99%

See 4 more Smart Citations

Pictogram Prediction in Alternative Communication Boards: a Mapping Study

Pereira

Medeiros

Zanchettin

et al. 2022

Anais Do XXXIII Simpósio Brasileiro De Informática Na Educação (SBIE 2022)

View full text Add to dashboard Cite

Alternative communication boards (ACB) are tools that try to compensate for the difficulties faced by people with complex communication needs. Generally, these tools consist of a mobile application in which the user can construct sentences by arranging pictograms (picture+label pair representing a concept) in sequence. This study systematically maps the literature on pictogram prediction in ACBs. We analyzed eight studies to investigate how computational methods are used for pictogram prediction, how these proposals are evaluated, and the studies' outcomes regarding user communication improvement. The main findings indicate the usage of different methods for pictogram prediction and a mixture of automatic and expert evaluation, which lead to inconclusive outcomes regarding user communication improvement.

show abstract

Section: Resultsmentioning

confidence: 99%