Luatodonotes: Boundary Labeling for Annotations in Texts

Abstract

We present a tool for annotating Latex documents with comments. Our annotations are placed in the left, right, or both margins, and connected to the corresponding positions in the text with arrows (so-called leaders). Problems of this type have been studied under the name boundary labeling. We consider various leader types (straight-line, rectilinear, and Bézier) and modify existing algorithms to allow for annotations of varying height. We have implemented our algorithms in Lua; they are available for download as an easy-to-use Luatex package. IntroductionMany word processing systems support annotations for the text. The most common case for this annotations are comments, which can be inserted in arbitrary positions inside the text. The comments themselves are placed as labels in the margin next to the text and connected to the corresponding position, called site, by a line called leader. The endpoint of a leader at a label is called a port. Such comments are available, for example, in LibreOffice (see Fig. 1) and Microsoft Word. This task can be expressed in the boundary labeling notion introduced by Bekos et al. [5]: the sites to be annotated lie inside the text area and the labels are to be placed outside the text area. They describe several types of leaders, such as straight-line leaders (s-leaders), rectilinear leaders with one bend (po-leaders) and rectilinear leaders with two bends (opo-leaders).Previous work. Boundary labeling has been extensively investigated in the last few years, see a survey on the interaction between cartography and graph drawing [17]. For labels of uniform size, the problem is well-studied. Most algorithms try to minimize the total leader length. For s-leaders, it suffices to compute a minimum-weight perfect matching, which can be done in O(n 2+ε ) time [1]. For opo-leaders, Bekos et al. [5] gave three different algorithms for the number of sides used by the labels, with running times O(n log n) (one-sided), O(n 2 ) (two-sided), and O(n 2 log 3 n) (four-sided). Further, they presented an O(n 2 )-time algorithm for po-leaders that lie on one side or on two opposite sides of the text. The result for po-leaders was improved by Benkert et al. [6] for the one-sided case. They gave an O(n log n)-time algorithm for length minimization and an O(n 3 )-time algorithm for a very general class of objective functions, including, for example,

External Labeling Techniques: A Taxonomy and Survey

“…First of all, most of the leaders used in external labeling algorithms are either straight-line leaders (see Section 4.1) or polylines whose segments have restricted slopes (see Section 4.2). Only few papers also considered curved leaders [KLW14]. The choice of the leader type has a strong influence on the overall appearance.…”

“…For straight-line leaders they considered uniform labels on all four sides of the rectangle and presented algorithms that find for n sites a plane labeling in O(n log n) time and a plane labeling minimizing the total leader length in O(n 2+δ ) time for any δ > 0; for details refer to Sections 5.2.4 and 5.2.2, respectively. Kindermann et al [KLW14] considered boundary labeling for straight-line leaders in the context of annotating texts. They placed the labels either in the left or right margin of the text.…”

“…The reason for the algorithm's much higher complexity compared to O(n 2 ) for length minimization [BKSW07] is that for non-length-minimal solutions the sites can no longer be easily split into two independent subproblems for the two opposite sides. The extension of Kindermann et al [KLW14] for non-uniform labels in 1-sided boundary labeling is not properly generalized to optimize 2-sided instances. Instead, for their text labeling application, the authors proposed to simply split the sites at the horizontal median position into two balanced parts and then run the 1-sided algorithm for each side independently.…”

“…Few works studied external labelings with polyline leaders that are neither orthogonal nor orthodiagonal. In text annotation, Kindermann et al [KLW14] proposed boundary labelings with os-leaders, where the first segment of each leader is used to reach the margin of the text, while the second one has a slope such that all labels can be arranged on the margin of the text. Obviously, such labelings can be derived from opo-labelings by removing the last bend of each leader.…”

“…Other labeling techniques such as dynamic programming (S5.4), meta-heuristic (S5.5), matchings (S5.6) and sweep-line algorithms (S5.7) are rarely used. Fekete and Plaisaint[FP99], Battersby et al [BSF * 11] Kindermann et al[KLW14], and Götzelmann et al[GHS06b] used highly specialized algorithms that do not match any of the listed techniques; details are given in Section 4.1.2.…”