The Wor(l)d Builders - Designing for the Improvement of Middle School Pupils Writing Practices Utilizing Natural Language Processing and Didactics Strategies
Authors
Kormind, Samuel ; Nielsen, Lukas Rasmus
Term
4. term
Education
Publication year
2018
Abstract
Dette speciale undersøger, hvordan man kan motivere mellemtrinselever til at skrive mere og bedre ved at kombinere spilbaseret læring, naturlig sprogbehandling (NLP) og didaktiske strategier. Vi designer og implementerer The Wor(l)d Builders, et narrativt computerspil udviklet i Unity, hvor elevernes primære interaktion er at skrive sætninger, som via afhængighedsparsing omsættes til objekter i spillets verden. Eleverne støttes gennem didaktisk modellering og tydelige skriveformål. Det foreslåede design og dets metoder blev afprøvet med mellemtrins- og gymnasieelever (n = 51) ved hjælp af en nøjagtighedsmåling for NLP-delen (67,4 %) samt spørgeskemaer og kvalitative svar. Resultaterne peger på, at afhængighedsparsing og didaktisk modellering er anvendelige i denne kontekst, mens Word2Vec i sin aktuelle form ikke understøttede formålet tilstrækkeligt; den målte nøjagtighed kan forbedres. Specialet bidrager med et design og en empirisk afprøvning, der indikerer potentialet i at bruge narrativ kontekst til at styrke elevers skrivepraksis.
This thesis explores how to motivate middle school pupils to write more and better by combining game-based learning, natural language processing (NLP), and didactic strategies. We design and implement The Wor(l)d Builders, a narrative computer game built in Unity in which writing full sentences is the main interaction; a dependency parser analyzes each sentence to derive word relations that are used to generate objects in the game world, while didactic modeling supports pupils’ sentence construction. The proposed design and its methods were tested with middle and high school pupils (n = 51) using an NLP accuracy measure (67.4%) alongside questionnaires and qualitative responses. Findings indicate that dependency parsing and didactic modeling are viable in this context, whereas Word2Vec underperformed for the thesis purpose; the measured accuracy can be improved. The work contributes a design and empirical evaluation suggesting the potential of narrative context to support pupils’ writing practices.
[This summary has been generated with the help of AI directly from the project (PDF)]
Documents