On the Frailty of Universal POS Tags for Neural UD Parsers

Anderson, Mark; Gómez-Rodríguez, Carlos

doi:10.18653/v1/2020.conll-1.6

Cited by 9 publications

(5 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A slightly more pressing limitation is the absence of a feature analysis because certain systems could potentially benefit from different features. Work has been presented in this direction and has shown that predicted POS tags aren't wonderfully useful Anderson and Gómez-Rodríguez, 2020b;Zhang et al, 2020b). However, these analyses didn't include SEQLAB parsers at all and the transition-based system used was a lowerperforming system, UUParser.…”

Section: Limitations Of Analysismentioning

confidence: 99%

A Modest Pareto Optimisation Analysis of Dependency Parsers in 2021

Anderson

Gómez-Rodríguez

2021

Proceedings of the 17th International Conference on Parsing Technologies and the IWPT 2021 Shared Task on Parsing Into Enhanced

Self Cite

View full text Add to dashboard Cite

We evaluate three leading dependency parser systems from different paradigms on a small yet diverse subset of languages in terms of their accuracy-efficiency Pareto front. As we are interested in efficiency, we evaluate core parsers without pretrained language models (as these are typically huge networks and would constitute most of the compute time) or other augmentations that can be transversally applied to any of them. Biaffine parsing emerges as a well-balanced default choice, with sequence-labelling parsing being preferable if inference speed (but not training energy cost) is the priority.

show abstract

Section: Limitations Of Analysismentioning

confidence: 99%

A Modest Pareto Optimisation Analysis of Dependency Parsers in 2021

Anderson

Gómez-Rodríguez

2021

Proceedings of the 17th International Conference on Parsing Technologies and the IWPT 2021 Shared Task on Parsing Into Enhanced

Self Cite

View full text Add to dashboard Cite

show abstract

“…The other main contender to alter would be the embedding layers. For example, we could have altered the size of the character BiLSTM/CNN, but certain experiments show that it has a limited impact on accuracy Anderson and Gómez-Rodríguez, 2020b).…”

Section: Methodsmentioning

confidence: 99%

Section: Limitations Of Analysismentioning

confidence: 99%

Proceedings of the 17th International Conference on Parsing Technologies and the IWPT 2021 Shared Task on Parsing into Enhanced Universal Dependencies (IWPT 2021)

2021

View full text Add to dashboard Cite

When learned without exploration, local models for structured prediction tasks are subject to exposure bias and cannot be trained without detailed guidance. Active Imitation Learning (AIL), also known in NLP as Dynamic Oracle Learning, is a general technique for working around these issues by allowing the exploration of different outputs at training time.AIL requires oracle feedback: an oracle is any algorithm which can, given a partial candidate solution and gold annotation, find the correct (minimum loss) next output to produce. This paper describes a general finite state technique for deriving oracles. The technique described is also efficient and will greatly expand the tasks for which AIL can be used.

show abstract

“…these books or three books or these three books vs. *three these books). The cost of these predictions is a richer (fine-grained) set of categories to be considered (much richer than UPOS tagset) and we are aware of the implication that this richer set might have on efficiency and robustness of various models training (Anderson and Gómez-Rodríguez 2020).…”

Section: Robustness Predictivity In Performance Data and The Utility ...mentioning

confidence: 99%

Parameters of Cross-linguistic Variation in Expectation-based Minimalist Grammars (e-MGs)

Chesi

2023

ijcol

View full text Add to dashboard Cite

The fact that Parsing and Generation share the same grammatical knowledge is often considered the null hypothesis (Momma and Phillips 2018) but very few algorithms can take advantage of a cognitively plausible incremental procedure that operates roughly in the way words are produced and understood in real time. This is especially difficult if we consider cross-linguistic variation that has a clear impact on word order. In this paper, I present one such formalism, dubbed Expectation-based Minimalist Grammar (e-MG), that qualifies as a simplified version of the (Conflated) Minimalist Grammars, (C)MGs (Stabler 1997(Stabler , 2011(Stabler , 2013, and Phase-based Minimalist Grammars, PMGs (Chesi 2005(Chesi , 2007Stabler 2011). The crucial simplification consists of driving structure building only using lexically encoded categorial top-down expectations. The commitment to the top-down procedure (in e-MGs and PMGs, as opposed to (C)MGs, Chomsky 1995; Stabler 2011) will be crucial to capture a relevant set of empirical asymmetries in a parameterized cross-linguistic perspective which represents the least common denominator of structure building in both Parsing and Generation.

show abstract

On the Frailty of Universal POS Tags for Neural UD Parsers

Cited by 9 publications

References 17 publications

A Modest Pareto Optimisation Analysis of Dependency Parsers in 2021

A Modest Pareto Optimisation Analysis of Dependency Parsers in 2021

Proceedings of the 17th International Conference on Parsing Technologies and the IWPT 2021 Shared Task on Parsing into Enhanced Universal Dependencies (IWPT 2021)

Parameters of Cross-linguistic Variation in Expectation-based Minimalist Grammars (e-MGs)

Contact Info

Product

Resources

About