Together, the vertical and horizontal juxtapositions are designed to help users identify both high-level commonalities and nuanced variations across structurally similar sentences.

We process this data in a three-stage pipeline (Figure 6). In the first stage, Sentence Segmentation and Categorization, abstracts are split into individual sentences using the NLTK package, and each sentence is classified into one of the five pre-defined aspects as listed in Section 4.1.1. Classification is performed by prompting an LLM (see prompt used in Appendix D.1) with the sentence and its full abstract.

Then, we segment sentences within each aspect into grammar-preserving chunks (see prompt used in Appendix D.2). This results in grammatically coherent chunks that are the basis of structure patterns. After identifying chunk boundaries, we again prompt an LLM to generate labels for chunks in a human-in-the-loop approach: starting from an initial set of labels for chunk roles, when a new label is generated, a researcher from the research team examines the new label and merges it with existing labels if appropriate, controlling for the total number of labels.

After obtaining an expanded set of high-level chunk labels, we assign them to each of the sentence chunks by using LLMs in a multiclass classification few-shot learning task, with the initial labels and assignment as examples (see prompt used in Appendix D.3).

Then, we segment sentences within each aspect into grammarpreserving chunks (see prompt used in Appendix D.2). This results in grammatically coherent chunks that are the basis of structure patterns. After identifying chunk boundaries, we again prompt an LLM to generate labels for chunks in a human-in-the-loop approach: starting from an initial set of labels for chunk roles, when a new label is generated, a researcher from the research team examines the new label and merges it with existing labels if appropriate, controlling for the total number of labels.

We conducted a qualitative analysis of user study transcripts and survey responses using a Grounded Theory approach [8]. First, the lead researcher collected a list of participants' behaviors, approaches, reflections on their experience, and feedback about the interface. The researcher then systematically coded this data, revisiting the data multiples times and refining the codes to ensure consistency and coherence. Through this process, high-level themes were identified and organized using affinity diagramming. Once the thematic structure was finalized, the researcher gathered supporting evidence for each theme and synthesized the findings, which were reviewed by the research team to ensure agreement on the results.

Interviews were video and audio recorded. We transcribed the audio using OpenAI's Whisper automatic speech recognition system and anonymized the transcript before analysis.

Interviews were video and audio recorded. We transcribed the audio using OpenAI's Whisper automatic speech recognition system and anonymized the transcript before analysis. We analyzed the interview data using thematic analysis [1]. First, two members of the research team independently coded four (25% of collected data) randomly chosen participant data to generate low-level codes. The inter-coder reliability between the coders was 0.88 using Krippendorff's alpha [37]. The two coders then met together to cross-check, resolve coding conflicts, and consolidate the codes into a codebook across two sessions. Using the codebook, the two coders analyzed six randomly selected participant data each. The research team then met, discussed the analysis outcomes, and finalized themes over three sessions.

We recruited 12 active researchers by word of mouth followed bysnowball sampling [ 53 ]

Future work could explore more seamless ways of preserving context, such as allowing users to navigate through every sentence of an abstract directly within the Cross-Sentence Relationship pane, fostering a more cohesive understanding of the content.

In this sense, AbstractExplorer enables dialectical activities that users may otherwise have found to be too tedious or difficult to engage with.

According to SMT, this generalization depends on most documents having some shared implicit structure.

Supporting a new corpus would require defining new corpus-appropriate predefined aspects and chunk role labels. Other components would likely be able to remain unchanged.

By definition, sensemaking and other dialectical activities necessitate engagement.

In this work, we introduce a new paradigm for exploring a large corpus of small documents by identifying roles at the phrasal and sentence levels, then slice on, reify, group, and/or align the text itself on those roles, with sentences left intact.

Our work demonstrates that designs informed by Structure-Mapping Theory can support users in navigating, making use of, and engaging with variation present in information. In this sense, AbstractExplorer enables dialectical activities that users may otherwise have found to be too tedious or difficult to engage with.

Like prior Structural Mapping Theory (SMT)-informed work in text corpora representation, AbstractExplorer's features have enabled some users to see more of both the overview and the details at the same time, facilitating abstraction without losing context.

Like prior Structural Mapping Theory (SMT)-informed work in text corpora representation, AbstractExplorer's features have enabled some users to see more of both the overview and the details at the same time, facilitating abstraction without losing context.

Dialectical activities cannot be done on a user's behalf by AI; with variation affordances, AI is supporting the user's engagement with the data themselves.

We posit that our approach can generalize to other domains such as journalism, code synthesis, and social media analytics where visual alignment of text can enable meaningful comparisons of underlying patterns to identify relational clarity.

In this work, we introduce a new paradigm for exploring a large corpus of small documents by identifying roles at the phrasal and sentence levels, then slice on, reify, group, and/or align the text itself on those roles, with sentences left intact.

Dialectical activities cannot be done on a user's behalf by AI; with variation affordances, AI is supporting the user's engagement with the data themselves.

The ablation and summative studies verified the value of Abstract-Explorer, specifically showing that all three components of the Structural Mapping Engine—color coding, sentence ordering, and vertical alignment—are crucial for facilitating comparative close reading at scale.

Our work demonstrates that designs informed by Structure-Mapping Theory can support users in navigating, making use of, and engaging with variation present in information.

We posit that our approach can generalize to other domains such as journalism, code synthesis, and social media analytics where visual alignment of text can enable meaningful comparisons of underlying patterns to identify relational clarity.

We demonstrate how slicing sentences according to roles and visually aligning them can help readers perceive cross-document relationships in a coherent manner.

Our work demonstrates that designs informed by Structure-Mapping Theory can support users in navigating, making use of, and engaging with variation present in information.

pre-computing and reifying cross-document analogous relationships make it psychologically possible for users to engage—if they are willing to be guided by it. (Lower NFC users are more likely to fall into this category.)

These observations underscored the need for an improved UI design for authoring custom aspects, especially for those who are less familiar with the research corpus.

AbstractExplorer’s features have enabled some users to see more of both the overview and the details at the same time, facilitating abstraction without losing context.

Activity log data, which revealed how participants actually used the interface, echoed the above findings. According to the log data, participants spent most of their reading time (66.31%) with vertical alignment on the second element in structure pairs, followed by alignment on the first element (29.19%), and left-justified alignment (5.13%). Highlighting usage showed a similar preference: 91.13% of time with all chunks highlighted, 8.25% with partial highlighting, and minimal time (0.63%) without highlights.

Immediately after the focused reading task, we conducted a short interview asking participants to reflect on their experience with both tasks (Appendix J.2), followed by a post-study survey (Appendix J.3).

In this section, we present findings on how AbstractExplorer supports comparative close reading at scale by integrating quantitative survey responses and log data with qualitative analysis of transcripts and open-ended responses. The qualitative analysis process is described in detail in Appendix H.

Throughout the two tasks, we also collected detailed interaction logs including counts of user-defined aspects created, duration of highlighting usage, and time allocation across the three possible alignment options.

After the ablation study validated the effectiveness of all three SMT-inspired features together (especially for lower NFC users), we completed the implementation of AbstractExplorer and eval-uated its impact on researchers’ reading and sensemaking of a corpus of all ∼1000 paper abstracts from ACM CHI 2024.

The most popular condition had all three features enabled, i.e., 11 out of 24 participants (≈ 50%) preferred Figure 7C, as shown in the “Preferred” columns of Table 1. The remaining participants were roughly evenly split between the no-features baseline (6 par-ticipants) and the without-alignment ablation condition (5 partic-ipants). One participant each liked the without-highlighting and without-ordering ablation conditions most, respectively.

Participants with lower NFC more frequently preferred and experienced less cognitive load when skimming with all the features enabled.

Using a two-tailed Mann-Whitney U Test, we found that participants who reported their lowest perceived cognitive load when all three features were enabled had significantly lower NFC than participants who reported their lowest cognitive load level when skimming with no features enabled—in the baseline interface (p=0.03).

Both gaze data and the semi-structured interviews revealed that lower NFC participants were more willing to be guided by the three features and took advantage of them consciously.

To compute a participant's NFC score, we averaged their response to the six questions, each ranging from 1 to 7, after reversing the appropriate questions.

For simplicity of analysis, we denote participants with NFC scores above the overall participants' median NFC of 5.42 (IQR = 0.583) as higher NFC, and lower NFC otherwise.

The study concluded with a 15-minute semi-structured interview. During the interview, participants saw screenshots from the three conditions and were asked which they preferred and disliked, why, what they wished the interface had, what influenced their skimming, and how they normally skimmed texts.

These results suggest that these three features lose their effectiveness when not used together.

Lower NFC participants were generally guided by emergent visual patterns created by the interactions between features, especially blocks of color spanning multiple sentences created when all three features are turned on.

The raw NASA-TLX score is the sum of all 6 NASA-TLX questions after reversing the appropriate questions.

The study concluded with a 15-minute semi-structured interview. During the interview, participants saw screenshots from the three conditions and were asked which they preferred and disliked, why, what they wished the interface had, what influenced their skimming, and how they normally skimmed texts.

The most preferred condition (all three features enabled) was tied with the baseline no-features-enabled condition for lowest reported cognitive load. Specifically, 11 participants reported their lowest raw NASA-TLX scores8 in the all-three-features condition, and a different 11 participants reported their lowest raw NASA-TLX scores in the baseline condition.

In this study, we allowed participants to experience views of same-aspect sentences (Section 4.1.1) with different combinations of highlighting, ordering, and alignment (as described in Section 4.1.2 and Section 4.1.4) enabled or not, in order to understand which and/or what combinations most effectively supported users' ability to skim and read laterally across documents.

We collected 80 sentences from our abstracts dataset labeled by our system as "Methodology/Contribution." Participants viewed the same 80 sentences in each condition—often with a different subset of sentences initially visible due to ordering changes—but only had two minutes to look at them in each condition.

The specific research questions for this study were: (1) How do highlighting, alignment, and ordering affect reading patterns, user experience, and cognitive load? (2) How do participants’ valuation of these features relate to their Need for Cognition? (3) Does each feature provide value on its own, or only in conjunction with one or more of the other two features?

To contrast participants' gaze patterns in each condition, we used a Tobii Pro Spark eye-tracker placed below the desktop monitor used by all subjects; Tobii Pro Lab software recorded each participant's gaze over time in each condition.

In this study, we allowed participants to experience views of same-aspect sentences (Section 4.1.1) with different combinations of high-lighting, ordering, and alignment (as described in Section 4.1.2 and Section 4.1.4) enabled or not, in order to understand which and/or what combinations most effectively supported users’ ability to skim and read laterally across documents.

Structural mappings between objects are part of the cognitive process of comparison according to the Structure-Mapping Theory [17], and juxtaposition can facilitate humans in recognizing particular possible structural mappings between objects [75].

Inspired by GP-TSM [24], AbstractExplorer first segments sentences into grammar-preserving chunks—segments that respect grammatical boundaries, i.e., an LLM judges that the sentence can be truncated at that chunk boundary without breaking the grammatical integrity of the preceding text. Each chunk is then classified by an LLM as having one of nine pre-defined roles, each of which has its own assigned color.

We consider common sequences of chunk roles to be alignable structures that could be used to support users in identifying structural similarities and differences across sentences in different abstracts, in line with Structure-Mapping Theory [17].

This ordering prioritizes dominant structural patterns (largest groups first) while exposing fine-grained variations (via length-sorted triplets), mirroring how humans compare sentences, if SMT is an accurate description in this domain of comparative close reading.

In SMT terminology, rendering and arranging according to corresponding chunks reify "commonalities in structure," while variation within corresponding chunks are "alignable differences" that users are predicted to notice.

AbstractExplorer classifies sentences into five pre-defined aspects common in CHI abstracts: Problem Domain, Gaps in Prior Work, Methodology/Contribution, Results/Findings, and Discussion/Conclusion.

SMT posits that visual alignment helps people perceive relational similarities and differences more clearly, thereby improving their ability to make meaningful comparisons and understand underlying patterns [28, 38, 47].

After the interviews, we analyzed the data using the process described in Appendix B

The prior SMT-informed tools in Section 2.3 for both code and natural language corpora suggest that the cognitive process of comparing texts may be no exception to the cognitive processes SMT predicts.

In the first part of the session, we asked participants about their strategies for selecting publication venues for their manuscript submissions, how they identify and synthesize information from venues, their approaches to writing manuscripts, and finally, the technology they have used to help with these processes, current technology shortcomings, and ideas for addressing these challenges.

In order to determine (1) the context in which we might offer novel views of scientific abstracts and (2) the intelligibility of various novel prototype designs for reifying cross-abstract relationships, we conducted a formative interview study with 12 active researchers (see Appendix A for participant information).

We used these mock-ups as design probes [31] to inspire ideation and elicit creative responses. Specifically, we asked participants to compare and contrast alternative mock-ups and reflect on how they could be used or improved to support their known or emerging synthesis and information-foraging goals.

The prior SMT-informed tools in Section 2.3 for both code and natural language corpora suggest that the cognitive process of comparing texts may be no exception to the cognitive processes SMT predicts.

After the interviews, we analyzed the data using the process described in Appendix B

In the second part of the session, we provided participants with mock-ups of possible reifications of cross-document relationships that might help them synthesize information across abstracts.

AbstractExplorer specifically leverages the shared structure in scientific abstracts to facilitate abstract corpus skimming and comparative reading at scale.

SMT provides a framework for understanding how humans compare two or more objects by finding common structural alignments between objects.

Sessions, which were held on Zoom, lasted 55 minutes on average. Participants were compensated with $15 USD.

The interview sessions were divided into two parts: an open-ended semi-structured interview about their backgrounds and practices, followed by feedback on a range of mock-ups, including novel reified relationships between analogous sentences in different abstracts (Figure 2).

SMT provides a framework for understanding how humans compare two or more objects by finding common structural alignments between objects.

The revealed variation within these analogous cross-document relationships can invite the user’s engagement.

This is the essence of comparative close reading, a dialectical activity that requires repeated deep engagement with the texts to reveal new insights.

Structural Mapping Theory (SMT) is a long-standing well-vetted theory from Cognitive Science that describes how humans attend to and try to compare objects by finding mental representations of them that can be structurally mapped to each other (analogies).

These examples of text-centric lossless techniques do not abstract away or summarize; they strategically re-organize and re-render the existing text to help enhance readers' own perceptual cognition, informed by Structural Mapping Theory (SMT) [17].

This SMT-informed approach, which AbstractExplorer shares, tries to give this mental machinery “a leg up,” letting users perhaps skip some steps by accepting reified cross-document relationships identified by the computer.

An ablation study (N=24) validated that these features work best together.

A summative study (N=16) describes how these features support users in familiarizing themselves with a corpus of paper abstracts from a single large conference with over 1000 papers.

The human perceptual, comparative mental machinery that SMT describes is part of what enables humans to form more abstract structured mental models from concrete examples, among other critical knowledge tasks.

AbstractExplorer instantiates new minimally lossy2 SMT-informed techniques for skimming, reading, and reasoning about a corpus of similarly structured short documents: phrase-level role classification that drives sentence ordering, highlighting, and spatial alignment.

Theory (SMT) to facilitate seeing both the overview and the details at the same time, facilitating abstraction without losing context.

Researchers have done rich historical investigations of individual notations (e.g., [3, 70, 74, 76, 115]), but the more general mechanisms and patterns through which new notations are created and formalized are less understood.

We began with a diverse set of notations across five disciplines—music, dance, chemistry, physics, and computer programming—that had prior historical literature to draw upon in our initial analysis.

Since histories of specific notations tends to miss detailed, direct observations around the initial creation process, we complement this "macro" analysis with occasional references to experiment-based literature from experimental semiotics, communication theory, and cognitive science into how people use notations to ground communication, largely in lab studies.

we conducted a comparative historical analysis of the development of different notations which individually have been documented in prior literature. Specifically, we conduct a parallel comparative history which "seek[s] above all to demonstrate that a theory similarly holds good from case to case... [and where] differences among the cases are primarily contextual particularities against which to highlight the generality of the [theorized] processes"

Studying software teams, Cherubini et al. [34] found a "tendency to adopt informal, ad-hoc notations" and a "limited adherence to standards of any sort."

Studies are also conducted on various existing notating practices, usually in specific domains (e.g., how programmers draw diagrams to communicate ideas [34, 62, 63]).

Studies are also conducted on various existing notating prac-tices, usually in specific domains (e.g., how programmers drawdiagrams to communicate ideas [34 , 62 , 63 ]).

What seem today as obvious notations often have relatively short histories: for instance, arrows in diagrams emerged around the 18th century.

Notations are deployed and embedded throughout the process of HCI and software development.

Almost everything we do with computers involves notations.

Recent advancements in AI offer a new paradigm of subjectively inferring human intent from ambiguous input [165].

Recent advancements in AI offer a new paradigm of subjectively inferring human intent from ambiguous input.

These informal interactions can then lead to formal representations, but depend upon pre-existing formalisms known to both humans and AI.

Given that user interfaces are a form of notation, it is not surprising that their evolution closely follows the patterns we identified.

Seemingly 'obvious' notations to academics are also not obvious to everyone: e.g., about one-third of the U.S. and German populations have low literacy in reading data visualizations [51].

As current AI technologies rely upon, reproduce, and amplify established, dominant, already-formalized abstractions and notations in order to function

Many notations are culturally learned and inherited.

In this work, we lay some groundwork towards addressing thisquestion by presenting a parallel comparative historical analy-sis [ 129] of notation development across scientific, computing,and artistic disciplines.

From our analysis, we derive a set of initial implications for the design of future systems that create new abstractions (Section 5), including that notations primarily originate through linking metaphors and most often in a social—rather than a technical—context, and that notation design decisions around what to include as "meaningful" (and thus what to exclude) are often left implicit by inventors, but could be made explicit and become manipulable objects through reification [10].

Our work contributes to a longstanding dream of dynamic abstractions in HCI, where users can dynamically communicate and express themselves through notations (interfaces) that they are most comfortable with at the moment of expression, beyond ones predefined by developers [96, 143, 144, 148, 149].

Here we present suggestions for system designers, with concrete examples inspired by our patterns. These are just some interesting ideas that came to mind, rather than an exhaustive list.

Alongside the social stages of notation development above are three functional stages that emerge from reflection upon our analysis—descriptive, generative, and evaluative stages (borrowing terminology from Generative Theories of Interaction [11])

Our historical analysis suggests that, cognitively and socially, a notation proceeds by: (1) Enumerating dimensions of meaningful variation in the target domain, which proliferate as more situations are encountered or considered (whether by inventors or users) (2) Mapping dimensions of meaningful variation to perceptual channels of representation (3) Designing the notation to leverage perceptual affordances by visual analogy to embodied transformations like pouring cups or rotating shapes, and ensuring these "natural" manipulations hold meaning in the target domain

These stages form a spectrum and are not rigid boundaries. We clustered patterns into the most relevant stage for ease of presentation; however, patterns can be applicable across stages.

Our review identified many empirical patterns in the notation development process. We state each pattern, briefly describe it, and provide examples.

Our work heeds calls from HCI scholars for more historicism in the field, to better contextualize technology "within dynamic temporal processes of emergence, change, continuity, decline, disappearance, or revival," and to see "the past... as a repository of design knowledge and experience"

From our analysis, we derive a set of initial implications for the design of future systems that create new abstractions (Section 5), including that notations primarily originate through linking metaphors and most often in a social—rather than a technical—context, and that notation design decisions around what to include as "meaningful" (and thus what to exclude) are often left implicit by inventors, but

Our analysis identifies 33 patterns of how notations are created, evolved, and formalized over time, which are largely shared across histories and loosely categorized into three social stages of development (invention/incubation, dispersion/divergence, and institutionalization/sanctification) and three functional stages (descriptive, generative, and evaluative).

What about novel formalisms and notations? How are new abstractions created, evolved, and incrementally formalized over time—and how might new systems, in turn, be explicitly designed to support these processes?

How might we co-create a new notation with a machine, and thereafter communicate through that notation, even share out the notation to broader communities?

While current AI systems support "horizontal" translations from informal ideas to established notations, how should we ensure that the "vertical" process of creation—new notations, new abstractions—is also supported?

Why and when is formalizationhelpful, and when is it not?

What stages does a notation go through until itis considered sufficiently “formalized”?

How is a notation initially created? Arethere common steps or techniques?

Why does a notation emerge? What com-pels or necessitates the creation of a new notation?

How do humans ultimately develop new notations, new formalisms, and new abstractions, that they use to communicate with machines and each other?

The use of notation happens everyday in small ways, e.g., whenever people work together over a whiteboard or paper towards a joint objective. People jot down X's, boxes and arrows to stand-for concepts they are working through.

Recent AI systems promise a new set of informal interactions withcomputers through natural language and other notational forms.

Human-computer interactions have historically been mediated by formally-defined structures—such as command-line interfaces, graphical user interfaces, and programming languages—that provide an unambiguous mapping to an underlying formal model.

Yet humans collaborating together do not always defer to a pre-existing formalism. Instead, they can develop ad-hoc, new notations to ground their communication [22, 34, 143], treating notations more like malleable resources than rigid systems.

Traditional human-computer interaction takes place through formally-specified systems like structured UIs and programming languages.

SDT broadly differentiates three types of motivation [157]: Intrinsic motivation denotes activity pursued for its inherently interesting or enjoyable qualities. Extrinsic motivation refers to activity pursued for a separable outcome. Amotivation denotes the absence of intentional motivation, where a person may no longer be aware why they pursue an activity.

Basic psychological needs theory (BPNT) posits three basic psychological needs that energise organismic processes: competence, the feeling of having an effect; autonomy, a sense that actions are self-endorsed and performed willingly; and relatedness, a sense of reciprocal care, value, and belonging in relation to other social figures and collectives [158].

SDT is broadly organised into six mini-theories, whose underlying concepts are continuously developed, critiqued, and revised (e.g., [186, 190, 191]).

At its core, SDT is a scientific theory [163], in that it contains a number of empirically-testable propositions [199] that generalise across varied contexts, which serve to explain and predict the impact of certain events on motivation and wellbeing.

SDT is a psychological macro-theory of human motivation, growth, and wellbeing [47, 48, 163] that characterises humans as fundamentally active organisms.

Self-Determination Theory (SDT), a major psychological theory of human motivation, has become increasingly popular in Human-Computer Interaction (HCI) research on games and play.

Self-Determination Theory (SDT), a major psychological the-ory of human motivation, has become increasingly popular inHuman-Computer Interaction (HCI) research on games andplay.

To our knowledge, the first SDT research involving videogames [18] was conducted shortly after Deci's original formulation of CET [129] and investigated whether extrinsic rewards would reduce intrinsic motivation even for 'highly intrinsically motivating' activities such as videogame play. Videogames' intrinsically motivating qualities were also examined in early research on learning [e.g., 351]; however, focused examination of other core SDT concepts such as need satisfaction largely began much later [365].

Research on games and play in HCI (henceforth HCI games research), however, has continued to employ broad psychological theories as foundational work [417, 556]. One prominent example can be seen in self-determination theory (SDT) [481, 483], an influential theory of human motivation, which has provided HCI games research with propositions and concepts that can help explain motivational and experiential qualities of games and game-adjacent systems (e.g., gamification).

Psychological concepts and models have long been employed in human–computer interaction (HCI) to theorise the human user [88]. However, early applications of cognitive psychological theory did not develop into a coherent foundation of knowledge about human factors [89, 109, 455]—circumstances that Rogers [456, p. 22] attribute to "the stark differences between a controlled lab setting and the messy real world setting" for which interactive artefacts and systems are designed. The deployment of broad theory in HCI has subsequently declined in the intervening years [455, 456], and this sporadic progress in theory development in domains such as usability and user experience (UX) has been identified as a cause for concern [249, 314].

Psychological concepts and models have long been employed in human–computer interaction(HCI) to theorise the human user [88].

In HCI, frameworks function as a type of theoretical contribu-tion, often supporting ideation, design, and evaluation.

A father tells his daughter, “It’s time to put the tablet down.” Not wanting to stop using the tablet, butworried about the consequences of disobedience, the child finds herself in a dilemma. With a strokeof insight, she puts the tablet down on the table in front of her, and keeps playing with it. She can stilluse the tablet, and her father’s instructions were met. Technically.

An inital version is implemented in an ObservableHQ notebookwhere sets of lede sentences can be input.

First,it utilizes a neuro-symbolic pipeline to support Variation Theory-based counterfactual data generation.

Users teach models theirconcept definition through data labeling, while refining their ownunderstandings throughout the process

Alignment is a bilateral process; it refers not only to AI acting according to human intentions but also to humans better leveraging AI by understanding the mechanisms behind it [54].

Data labeling as a cognitive task—including defining a concept or determining how two similar objects may have different labels—requires both comparison and integration [62].

However, relying exclusively on existing examples is not ideal for tasks requiring nuanced understanding of user intentions, as these examples often fail to represent diverse and edge-case scenarios [31].

When training samples are scarce, model performance heavily depends on the quality of available training examples [15].

An important challenge in interactive machine learning, particularly in subjective or ambiguous domains, is fostering bi-directional alignment between humans and models.

In supervised and semi-supervised machine learning (ML) pipelines, labeled data is a vital component of training and validating models [46].

In the context of co-adaptive learning, supporting the intertwined evolution of both the user's understanding and the model's learning is crucial [16].

Machine teaching, a part of the human-in-the-loop approach, has been used as a process in which a human expert (the "teacher") provides guidance to a machine learning model to help it learn important and robust features for decision making [57].

A targeted approach in IML is machine teaching (MT) [60], an interactive framework that allows users to devise and select useful data for labeling, with the goal of teaching the model relevant features during training [7, 18].

Interactive ML (IML) methods, like active learning [3], continuously apply human feedback during model training to iteratively build and refine the model [35, 42, 43].

Reflection is recognised as vital for rigorous and responsible (HCI)research, yet it is often treated as secondary, hidden in the marginsof papers.

The goal of this meet-up is to create a space for CHI attendees to discuss and practise reflection in HCI research and design.

This meet-up invites CHI attendees to come together inthe META HCI community to explore how we as researchers andpractitioners reflect on our own practices – and how we might doso more intentionally.

Reflection is not limited to one subfield or methodology: it is a concern that cuts across the entire discipline.

We understand reflection as a multifaceted concept [6, 24, 25] with implications and relevance at different levels for researchers [2, 3, 8, 28]:

Reflection has been a recurring theme in HCI – from Schön's reflective practitioner [24] to Sengers et al.'s reflective design [25].

Importantly, reflection happens on multiple levels: as individuals questioning assumptions and choices, as groups working together in projects or labs, and as a community negotiating shared values, norms, and directions.

Reflection, as we envision it here, is not limited to standardised methods and codified practices, but also includes the hidden, vulnerable conversations about academic life.

Structures — reflection on the structures that condition HCI and our own standings within them: societal constructs (positions, values, power) shaping what problems are visible and whose knowledge is legitimised.

Practice — reflection on practice (processes): the ways we design, study, and collaborate [32].

Self — reflection on the self: a deliberate, structured evaluation of one's thoughts, feelings, and actions [7, 13].

We understand reflection as a multifaceted concept [6, 24, 25] with implications and relevance at different levels for researchers [2, 3, 8, 28].

While reflection is acknowledged as crucial for rigorous and responsible research, it often remains tacit and under-discussed.

Reflection has been a recurring theme in HCI – from Schön's reflective practitioner [24] to Sengers et al.'s reflective design [25]. However, it is seldom centred in our collective conversations [2].

As the community around augmented reading broadens and as possibilities continue to unfold, it is the purpose of this workshop to set up our community to drive innovation in a productive, desirable, and responsible way.

The landscape of technology for consuming information is changing rapidly. One mode of information consumption, reading, stands to see profound changes due to its ubiquity and frequency as a cognitive task.

Recent changes in the technological landscape are significantly changing the reading experience. AI has introduced many new possibilities for interfaces to augment or transform text to be more rapidly scanned, navigated, understood, and compared to other texts.

Reading is one of the most ubiquitous modes for consuming information.

Reading is one of the most ubiquitous modes for consuming in-formation.

Visualization is central to scientific discovery, yet authoring tools remain split between information and scientific visualization,and expertise in one rarely transfers to the other.

Butanalogy can also operate in mutual alignment1 analogies to reveal commonalities thatwere previously not obvious in either analog.

Projecting information from a well-understood domain can lend structure to an unfamiliar domain, as in:The mitochondria are the power supply for a cell.

Analogy is often thought of chiefly as a way to transfer knowledge from one situationto another, and indeed, it often serves that function.

We illustrate ourpoints with examples from adults and children, including examples from language evolu-tion, and across both perceptual and conceptual domains.

We propose that—bothin the history of language and in children’s learning—analogical processes are a majorway in which new relational abstractions are acquired

But the ear-lier we go in development, the less able children are to comprehend verbal explanationsof abstract ideas. In contrast, there is evidence that analogical comparison and abstractionprocesses are present in 7–9-month-old infants, and even earlier (Anderson, Chang, Hes-pos, & Gentner, under review; Ferry, Hespos, & Gentner, 2015).

Relational categories have been the focus of much recent research (Asmuth &Gentner, 2017; Gentner, 2005; Gentner & Kurtz, 2005; Goldwater & Markman, 2011;Markman & Stilwell, 2001; Ross & Murphy, 1999), in part because of their importantrole in conceptual learning and education (Goldwater & Schalk, 2016).

For example, carnivore andherbivore are abstract relational categories, while canine and feline are abstract entity cat-egories.

Relational cate-gories are categories for which the basis for membership is participation in a commonrelational structure; thus, they differ from the more studied entity categories, such as tulipand spoon, whose members share many intrinsic properties.

Our main focus is on relational abstractions, includingprinciples, rules, and schemas, as well as abstract relational categories.

For example, causal system is more abstract than posi-tive feedback system, which in turn is more abstract than the specific positive feedbacksystem by which the melting of polar ice causes lower reflectance of the sun’s heat, lead-ing in turn to more rapid melting.

Wetake the process of abstraction to be one of decreasing the specificity (and therebyincreasing the scope) of a concept.

Many such abstractions are expressed as rela-tional categories—categories like evidence, counterfactual, and proportion, and on a moremundane level, bargain, ally, and rescue.

Theassertions that make up abstract knowledge are variously referred to as schemas, rules,abstractions, principles, or overhypotheses

Abstract structured knowledge is a key feature of higher order cognition (Gentner &Medina, 1998; Hummel, 2011; Markman, 1999; Tenenbaum & Griffiths, 2001).

it is not enough to consider the distribution of examples given to learn-ers; one must consider the processes learners are applying

Wepropose that analogical generalization drives much of this early learning and allows children togenerate new abstractions from experience

contrary to the general assumption,maximizing variability is not always the best route for maximizing generalization and transfer

Gebreegziabher et al. [24] argued that counterfactual generation that follows the principles of VT allowed the introduction of discriminatory variance for the model to learn on.

Building on methods proposed in PaTAT [24], Mocha first generates human-readable neuro-symbolic pattern rules from partially labeled text data for classification.

These theories have proven insightful for understanding how humans grasp and compare concepts, shaping the development of human-AI collaboration systems for sensemaking [29], hypothesis testing [2], as well as model training [24].

Both systems enabled users to quickly identify variations and patterns within the data and support exploration and hypothesis testing.

The last two prior works also combine Variation Theory (VT) and SAT together, as we did (i.e., a corollary of SAT referred to as Analogical Transfer/Learning Theory).

In line with previous work, Mocha aims to support a user's efforts in the disambiguation of concepts through structural comparisons of counterfactual data in the context of machine teaching.

Engineering refers to the use of technical principles, such as mathematics, science, and technical know-how, to realize a design that best meets a given set of expectations, which are typically captured in a requirements specification.

Designing is the process of arriving at a plan, specification, prototype, system, or service—a design. In HCI, this often means designing a user interface and relevant parts of the underlying interactive system.

HCI focuses on people who use an interactive system or are affected by its use. This focus is often called being user-centered or human-centered to contrast it with a focus on the technology itself [423, 604].

Finally, interaction often involves co-adaptation between people and computers [646], meaning that both the user and the system learn and adapt to each other during interactions.

Interaction is, in other words, not a property of the system design or the user but something that emerges when they influence each other.

The development of technology for interactive computing systems has been an important driver behind the widespread adoption of computing we have witnessed in the last 50 years.

In HCI, evaluation refers to the application of some systematic methodology to attribute human-related values to an artifact, prototype, system, or process. Examples of such attributes include performance, experience, safety, and ethical aspects, such as the avoidance of bias or harm.