Language Scent: Exploring Cross-Language Information Navigation

Abstract

While multilingual users often switch between languages when seeking information, this process remains undersupported by current systems where information is typically siloed by language. Our formative study reveals that users' cross-language transitions are guided by their perceived value of switching to a language, a concept we formalize as language scent. Language scent extends Pirolli and Card's theory of information scent to multilingual scenarios by considering meta-level strategy formation when navigating between different languages. To support language scent, we designed Niffler, a search system that augments language scent and supports cross-language information navigation through contextual cues, in-situ tools, and reflection support. A lab study with 16 multilingual speakers showed that Niffler facilitated the formation and execution of exploratory and granular search strategies and leads to diverse information being gathered. Our findings establish language scent as a valuable lens on cross-language information seeking, highlighting language's role in enabling access to broader information and offering concrete implications for the design of multilingual search systems.

Figures (7)

• Figure 1: System Overview. Niffler consists of [A] a search results page enhancement of [A1] Comparative Summary and Query Information, and [A2] Cross-Lingual Keywords, [B] in-situ tools, and [C] a side panel with tabs [C1] analysis and [C2] saved content.
• Figure 2: On the search results page, the [A] Comparative Summary consists of [A1] Cross-Lingual Comparison summarizing key similarities and differences between the two languages, and [A2] Summaries of English sources and Chinese sources, respectively. [B] Query Information is provided as background information to complement it. Each search result is decorated with [C] Cross-Lingual keywords.
• Figure 3: The tooltip opens when text is selected. Users can choose to see [A] Contextual Translation of the text, connecting to relevant user search history, or [B] Preview of Other Language, with suggested queries and sources in the other languages.
• Figure 4: The side panel analysis tab provides [A] language-centred visualizations of user search activity and [B] analysis functions to help them consolidate information and evaluate past strategies. There are two visualizations, the [A1] Timeline highlighting language switching patterns and the [A2] Semantic Tree focusing on high-level concepts. Users can select nodes on the visualization for analysis, being able to [B1] Summarize, [B2] Compare, or [B3] Suggest queries.
• Figure 5: The Baseline consists of [A] Parallel Search Panel and [B] AI Chat Panel.