Rationale Essay
As I prepare for a minor in digital linguistics, I wanted to design a fun, creative project around concepts in Natural Language Processing and Linguistic Theory that had challenged me, in order to deepen my understanding of complex theories. When reading Idioms by Nunberg et al. (1999), I was not expecting to read an argument suggesting the compositionality of idioms, as I had the same idea that Chomsky (1989) had addressed, suggesting that idioms were stored as whole units in our mental lexicons. However, after reading the article, I did understand the reasoning behind how some idioms can be more transparent than others, indicating that there is a degree of transparency among them. Grappling with these complex ideas, along with being introduced to key theories of statistical methods in NLP through Universal Dependency tools, I was inspired to create a project which would be the beginning base for mapping key theories of linguistics onto NLP.
Initially, I had thought to base my website on a single theory, but after a consultation with my professor I decided to acknowledge key theories and their significance, but not point to a single theory, addressing how no one theory is completely accurate or most effective for classifying multi-word unit expressions. However, throughout this project, I mainly focused on theories of Construction Grammar and Compositionality, briefly addressing Typology and Language Packaging, as well as Usage theory and Dependency Grammar.
One of the main struggles in computer science today, Multi-word units (MWEs) prove to be a “pain in the neck” for Natural Language Processing (NLP) (Sag et al., 2002), lying at the very heart of effective NLP (Constant et al., 2017). That being said, there are still many ongoing debates over how these expressions should be processed in NLP today (Villavicencio & Idiart, 2019). Additionally, MWEs have been proved difficult to define, as Rayson (2010) notes that “while there is no universally agreed definition for MWE as yet, most researchers use the term to refer to those frequently occurring phrasal units which are subject to certain level of semantic opaqueness, or non-compositionality” (p. 1). This project aims to compare key theories in linguistics regarding multi-word units, define key terminology, and relate them to NLP tools used today.
In conclusion, my project shows that no single linguistic theory is able to fully capture the complexities of multi-word unit expressions. Where Construction Grammar excels in dividing the words into individual units, it over-focuses on form–meaning connections when a phrase is non-compositional or opaque. On a similar note, compositionality can be used to explain degrees of transparency within idioms (Nunberg et al., 1999), but falls short in providing any insight into how to analyze non-compositional, phrasal idioms. For future research implications, my study indicates the need for multi-theory approaches to NLP.
*AI Transparency Statement*
In order to build my website layout, I asked ChatGPT for ideas on coding themes and how to apply them in WordPress using CSS and HTML. The outlines of the code themes allowed me to conduct more research on the main theories and function of the website, while getting a theme that matched my project. I was also able to learn how to use the code editor more effectively, as I was not aware of what it was able to do from my last projects.