Locality and transferability: examining pre-built lexicons to elicit landscape values from natural language

Abstract. Landscape values are perceived through complex interactions between people and their surroundings, and understanding such values is essential for policy. Previous research to explore landscape values through text analysis often relies on developing lexicons, which serve as value classification rules. However, it is unclear how transferrable such lexicons are between locations of differing environmental and cultural conditions.

In this work, we examine the transferability of lexicons from previous studies: one on based on Geograph data to contain natural language in the UK and another on based on TripAdvisor in the context of US national parks. Both lexicons have typologies for 1) attractiveness/aesthetics of landscape and 2) natural elements/mammal species/biological values of landscape. We apply these lexicons to a text corpus built with the Guardian’s Country Diary.

Our initial findings were spatial distributions of lexical matches, along with match ratios and keywords at county level in the UK. Then we zoomed in to compare the lexical performance to learn that larger lexicons do not guarantee better results when the context mismatches. Indeed, there is a room for transferability of lexicons; nonetheless, it is crucial to acknowledge that lexicons are sensitive to locality, urging to consider site-specific biophysical and sociocultural difference in applying pre-built lexicons.