EmphasisChecker: A Tool for Guiding Chart and Caption Emphasis

TL;DR

This work tackles the problem that readers' takeaways from charts and captions degrade when emphasis is misaligned. It introduces EmphasisChecker, a tool that concurrently highlights visually prominent features in time-series charts and the features emphasized in captions, using a time-series $\\varepsilon$-persistence detector based on the Ramer-Douglas-Peucker algorithm and a text-reference extractor built on Stanford CoreNLP/SUTIME and BERT embeddings. Across real-world data and a dedicated user study, the authors demonstrate that alignment between chart and caption emphasis is common but imperfect, with professional authors showing a 65% match rate and Tableau Public captions being mostly basic. The results indicate EmphasisChecker is useful and easy to use for authoring aligned chart-caption pairs, with potential to guide revisions, reduce misinterpretations, and inform broader extensions to other chart types and accessibility needs.

Abstract

Recent work has shown that when both the chart and caption emphasize the same aspects of the data, readers tend to remember the doubly-emphasized features as takeaways; when there is a mismatch, readers rely on the chart to form takeaways and can miss information in the caption text. Through a survey of 280 chart-caption pairs in real-world sources (e.g., news media, poll reports, government reports, academic articles, and Tableau Public), we find that captions often do not emphasize the same information in practice, which could limit how effectively readers take away the authors' intended messages. Motivated by the survey findings, we present EmphasisChecker, an interactive tool that highlights visually prominent chart features as well as the features emphasized by the caption text along with any mismatches in the emphasis. The tool implements a time-series prominent feature detector based on the Ramer-Douglas-Peucker algorithm and a text reference extractor that identifies time references and data descriptions in the caption and matches them with chart data. This information enables authors to compare features emphasized by these two modalities, quickly see mismatches, and make necessary revisions. A user study confirms that our tool is both useful and easy to use when authoring charts and captions.

Figures (9)

• Figure 1: As the author writes a caption about the chart (c) in the textbox (d), EmphasisChecker shows the chart's visually prominent features in (b) and the text references to the chart features (a). The interface shows (b) visually prominent chart features (unmatched features in orange $\blacksquare$ and matched features in green $\blacksquare$, marks above the chart). It uses circles to depict point features (e.g., local extrema; the peak around 1981) and bars to depict trend features (e.g., the rising trend up to 1981). In addition, it shows (a) references between the chart and the text (i.e., blue $\blacksquare$, red $\blacksquare$, purple $\blacksquare$, and brown $\blacksquare$ marks at the top of the page and on the text). In the input text box (d), the tool adds a red squiggly underline () on the phrase 'soared from 1980 to 1991', a typo of 'soared from 1980 to 1981' because the phrase does not match the data in the chart. The tool also adds a blue squiggly underline () on the phrase 'dip between 2008 and 2012' because the phrase does not match any of the prominent chart features.
• Figure 2: Distribution of chart and caption emphasis in chart-article pairs. Professionals often match chart and text emphases (blue $\blacksquare$ segment in the top bar) but occasionally do not (orange $\blacksquare$ (Unmatched: NP (non-prominent); captions only describing non-prominent chart features) + gray $\blacksquare$ (Unmatched: basic; basic captions that do not point to specific chart features) segments in the top bar). On the other hand, captions on Tableau Public are predominantly basic captions (gray $\blacksquare$ segment in the bottom bar). The values have been rounded to the nearest whole numbers and may not sum to 100%.
• Figure 3: Views of the EmphasisChecker interface from the usage scenario. The chart shows the real home price index between 1890 and 2006. (a) Prominent features are shown on top with a basic caption not describing any specific feature. (b) Caption text matches the most prominent visual feature (sharp rise on the right; blue $\blacksquare$ highlight in the UI). (c) Typo in the caption text indicated by a red squiggly underline () on 'declined since 1984'. (d) Caption text matching a less prominent feature, indicated by a blue squiggly underline () on 'declined since 1894.'
• Figure 4: EmphasisChecker tool overview. A user first uploads time-series data (including axis ranges and aspect ratio). Based on the input, the time-series prominent feature detector detects the visually prominent features and displays them to the user (Section \ref{['sec:prominent-detector']}). The user can type text based on the prominent features they see in the chart. When triggered, the text reference extractor identifies the references between the chart and text and displays them to the user, with comparisons with the prominent features (Section \ref{['sec:reference-identifier']}).
• Figure 5: EmphasisChecker's interface for authoring charts and captions. (a) The interface includes two switches, the chart edit mode for toggling the sliders (Component (b)) and the emphasis display mode that toggles the display of references (Figure \ref{['fig:teaser']}a) and the visually prominent features (Figure \ref{['fig:teaser']}b). The figure shows the state with chart edit mode on and emphasis display mode off. (b) When the chart edit mode is on, the user can manipulate the vertical and horizontal sliders to edit the dimensions of the chart and the axes ranges. (c) The user can hover over the chart to view tooltips showing the underlying data values. (d) The user can type the caption in the textbox.