A comparative analysis of deep learning and machine learning for POS tagging

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F26%3APT6ZLZ3T" target="_blank" >RIV/00216208:11320/26:PT6ZLZ3T - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1016/j.eswa.2025.128026" target="_blank" >http://dx.doi.org/10.1016/j.eswa.2025.128026</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.eswa.2025.128026" target="_blank" >10.1016/j.eswa.2025.128026</a>

Jazyk výsledku

angličtina

Název v původním jazyce

A comparative analysis of deep learning and machine learning for POS tagging

Popis výsledku v původním jazyce

Natural Language Processing (NLP) has experienced substantial alteration with the emergence of deep learning (DL), which increasingly prefers end-to-end architectures over conventional pipeline methodologies (for example, tokenization, POS tagging, and parsing). While Part-of-Speech (POS) labeling was formerly essential for syntactic parsing and downstream activities such as early machine translation, newer systems often avoid explicit POS annotation. Nonetheless, POS tagging remains relevant in particular contexts: (1) facilitating syntactic analysis in hybrid NLP systems, (2) providing linguistic scaffolding for low-resource languages where data-hungry DL models fail, and (3) enhancing interpretability in grammatical annotation tasks. This article gives a thorough assessment of POS tagging strategies from 2019 to 2023, including rule-based, statistical, machine learning, and deep learning techniques. We investigate their technological growth, capabilities, and limits, with a focus on their incorporation into modern NLP processes. Our results highlight new trends, such as parameter-efficient tagging for multilingual situations, as well as ongoing issues in morphologically rich languages. This paper highlights the transitory function of POS tagging within current NLP paradigms, which is lessened in general-purpose applications but persists in specialized areas, and reveals synergies between modular language analysis and end-to-end systems. © 2025 Elsevier Ltd

Název v anglickém jazyce

A comparative analysis of deep learning and machine learning for POS tagging

Popis výsledku anglicky

Natural Language Processing (NLP) has experienced substantial alteration with the emergence of deep learning (DL), which increasingly prefers end-to-end architectures over conventional pipeline methodologies (for example, tokenization, POS tagging, and parsing). While Part-of-Speech (POS) labeling was formerly essential for syntactic parsing and downstream activities such as early machine translation, newer systems often avoid explicit POS annotation. Nonetheless, POS tagging remains relevant in particular contexts: (1) facilitating syntactic analysis in hybrid NLP systems, (2) providing linguistic scaffolding for low-resource languages where data-hungry DL models fail, and (3) enhancing interpretability in grammatical annotation tasks. This article gives a thorough assessment of POS tagging strategies from 2019 to 2023, including rule-based, statistical, machine learning, and deep learning techniques. We investigate their technological growth, capabilities, and limits, with a focus on their incorporation into modern NLP processes. Our results highlight new trends, such as parameter-efficient tagging for multilingual situations, as well as ongoing issues in morphologically rich languages. This paper highlights the transitory function of POS tagging within current NLP paradigms, which is lessened in general-purpose applications but persists in specialized areas, and reveals synergies between modular language analysis and end-to-end systems. © 2025 Elsevier Ltd

Druh

J_SC - Článek v periodiku v databázi SCOPUS

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

—

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

Expert Systems with Applications

ISSN

0957-4174

e-ISSN

—

Svazek periodika

288

Číslo periodika v rámci svazku

2025

Stát vydavatele periodika

US - Spojené státy americké

Počet stran výsledku

25

Strana od-do

128026

Kód UT WoS článku

—

EID výsledku v databázi Scopus

2-s2.0-105006676145