Research Report: Design and Specification of the Parallel Reader and Concordance Workbench

Full report

On this page

Introduction and Epistemological Framework

Within the highly constrained, browser-based environment of the Al.Qaeda.net Archival Reconstruction, the preservation of exact, hash-verified canonical source identities forms the unalterable foundation of all subsequent historical analysis. Canonical sources—spanning preserved research documents, extracted text streams, and derived archive exhibits—must never be silently edited, normalized, or structurally compromised. However, the isolation of single documents in solitary viewer instances fundamentally limits the capacity for rigorous, cross-document comparative analysis. To bridge the critical gap between static preservation and active scholarly inquiry, the system requires a robust, local-first comparative reading instrument. This report outlines the comprehensive design, theoretical underpinning, and precise user interface mechanics for the recommended feature: the Parallel Reader and Concordance Workbench. Rooted deeply in digital humanities methodologies, this feature allows researchers to inspect two or more records simultaneously within a unified, historically authentic workspace. The workbench is designed strictly as an instrument for reading, alignment, and comparison. Its ultimate purpose is to illuminate textual variance, shared rhetoric, and structural divergence without declaring any single text as the "correct" or base witness. Implementing this functionality elevates the archival reconstruction from a passive repository or visual novelty into a serious research environment capable of supporting stemmatology, genetic criticism, and discourse analysis. By mapping the evolutionary differences across document versions, translation drafts, or ideological manifestos, researchers can trace the historical development of concepts and the precise manipulation of language over time1. The architecture detailed herein relies on period-authentic visual paradigms—specifically the classic Windows Multiple Document Interface (MDI)—preserving the nostalgic workstation aesthetic while providing advanced algorithmic collation running entirely within the local browser environment.

Historical Precedents and Scholarly Benchmarks

To ground the workbench in proven textual analysis methodologies, this design synthesizes principles from historic and modern scholarly instruments. The evolution of textual collation—from physical opto-mechanical devices to sophisticated global alignment algorithms—provides a vital blueprint for handling natural language variance.

These benchmarks demonstrate a vital transition in textual criticism: moving away from manual side-by-side reading (which is prone to human error and cognitive fatigue) toward algorithmic sequence alignment2. However, unlike code-oriented diff tools that aggressively highlight every byte difference, humanities-focused instruments like Juxta and CollateX are designed to tolerate the ambiguities of natural language, punctuation fluctuations, transcription anomalies, and Optical Character Recognition (OCR) errors8. The Parallel Reader and Concordance Workbench synthesizes the visual immediacy of WinDiff with the algorithmic sophistication of CollateX and the statistical power of AntConc.

Visual Architecture: The Multiple Document Interface (MDI)

The Parallel Reader and Concordance Workbench must seamlessly integrate into the recovered workstation aesthetic. Modern Integrated Development Environment (IDE) paradigms—such as deeply nested docking panels, flat design, or borderless floating tabs—are strictly prohibited. Instead, the visual architecture relies entirely on the classic Windows Multiple Document Interface (MDI) specification19.

The MDI Paradigm and Window Containment

In the MDI model, a single primary application window (the MDI Parent or Client Window) serves as a rigid visual and logical container for multiple nested child windows. Each child window within this application represents a distinct, encapsulated Comparison Session. The fundamental rule of MDI is strict geometric containment: child windows cannot be dragged outside the bounds of the MDI client area20. If a user attempts to move a comparison session beyond the parent window's borders, the operating system (or in this case, the browser-based window manager) clips the child window precisely at the parent's inner edge. When an MDI child window is maximized, its independent title bar disappears, merging its title text with the parent application's title bar. Simultaneously, standard Minimize, Restore, and Close control buttons are injected into the far right side of the parent application's menu bar20. This behavior maximizes screen real estate for deep reading while adhering to period-accurate window management protocols.

Z-Order Management and Rendering

The system maintains a strict Z-order for stacking overlapping windows20. Clicking anywhere on an inactive comparison window brings it immediately to the foreground, triggering a synchronous repaint event. To prevent visual artifacting (often seen in legacy systems where dragging a window leaves "trails"), the workbench utilizes off-screen bitmap caching. The rendering engine caches the bitmap of the objects under the active form, ensuring that when the comparison window is dragged across the screen, the underlying archival desktop repaints smoothly without forcing expensive DOM recalculations for every pixel of movement22.

Window Menu Controls and Spatial Topologies

The primary menu bar of the MDI application contains a dedicated "Window" dropdown that manages the spatial arrangement of the workspace. Standard MDI commands manipulate the geometry of all open child windows programmatically20:

• Cascade: Arranges all open MDI child windows in an overlapping, diagonal format. The windows overlap one another from the top-left to the bottom-right of the client area, but the title bar of each remains completely visible and clickable20.
• Tile Horizontally: Splits the available client area into equal horizontal bands (stacked rows).
• Tile Vertically: Splits the client area into equal vertical columns (side-by-side panes).

Comparison Layouts and Viewport Configuration

For reading and comparing natural language texts, the orientation and number of panes critically impact the user's cognitive load and the efficacy of the analysis.

Top/Bottom vs. Side-by-Side Modes

• Side-by-Side (Vertical Tiling): This is the optimal mode for prose, transcripts, and long-form book reading. It allows the researcher's eyes to track horizontally between aligned paragraphs. Side-by-side panels represent the traditional, historically validated mechanism for reading variant texts, mirroring the physical layout of bilingual bibles or facing-page translations24.
• Top/Bottom (Horizontal Tiling): This mode is better suited for inspecting short, highly structured records (such as database rows or poetry) or for concordances where the horizontal width is absolutely required to display extensive Keyword-in-Context (KWIC) strings without aggressive word-wrapping.

Optimal Number of Simultaneous Panes

Algorithmic alignment becomes visually and computationally chaotic beyond two or three witnesses. While the MDI environment theoretically allows infinite window spawning, a single Comparison Session should support a maximum of four simultaneous text panes. Two panes provide the ideal side-by-side reading experience for standard comparative analysis; four panes approach the absolute limit of readability on period-authentic screen resolutions (e.g., 1024x768 or 1280x1024).

Line Numbering and Paragraph Anchors

In a classic MDI text viewer, navigation and reference rely heavily on spatial markers. The workbench implements a classic gray left-hand "gutter" in each document pane. Users can toggle this gutter to display continuous Line Numbering or Paragraph Anchors. Because the texts are dynamically tokenized and may wrap differently depending on the pane's pixel width, Line Numbering is calculated based on the logical document lines (delineated by hard returns), not the visual wrapped lines. Paragraph anchors provide static reference points (e.g., ¶ 12\) that remain consistent regardless of window resizing, facilitating easy cross-referencing in exported reports.

Algorithmic Collation and The Gothenburg Model

To ensure the workbench serves as a serious research instrument rather than a superficial text viewer, the textual comparison engine is modeled precisely on the "Gothenburg Model." This architectural framework is the prevailing standard for modern scholarly collation8. The Gothenburg Model explicitly divorces the computational alignment logic from the final visual rendering, processing texts through a pipeline of discrete, consecutive steps8.

1. Tokenization

Before any comparison can occur, the raw text streams must be segmented into discrete, atomic units known as tokens. Tokenization typically splits strings at whitespace characters7. However, in the workbench, tokenization rules must treat punctuation separately from alphanumeric characters. For example, the string operation. must be tokenized into \[operation\] and \[.\]. If punctuation remains attached to the word, minor typographical differences (e.g., a missing comma) will cause the algorithm to register a massive substantive variant, severely degrading the quality of the alignment25.

2. Normalization and Regularization

Because the exact canonical sources within the archive cannot be permanently edited or altered, the comparison engine applies normalization strictly in memory during the collation phase. Normalization involves ignoring non-substantive variations to isolate true semantic differences8. Through the UI, users may toggle specific normalization rules:

• Ignore Case: Converts all tokens to lowercase before comparison.
• Ignore Whitespace: Collapses multiple spaces and tabs into a single generic separator.
• Ignore Punctuation: Strips all punctuation tokens from the comparison array.
• Heuristic OCR Correction: Applies a lightweight equivalence table to ignore common OCR artifacts (e.g., treating f and ſ (long s) as identical, or equating rn with m in degraded scans).

This in-memory regularization ensures the underlying, hash-verified text remains pristine while the comparative view filters out visual noise8.

3. Alignment via Needleman-Wunsch

The core of the comparison engine utilizes the Needleman-Wunsch algorithm, a dynamic programming technique originally developed in bioinformatics for DNA sequence alignment, which has been widely and successfully adapted for global sequence alignment in textual scholarship7. Needleman-Wunsch evaluates the entirety of both text sequences to find the optimal global alignment. It achieves this by creating a two-dimensional matrix where Document A forms the X-axis and Document B forms the Y-axis. The algorithm scores every possible path through the matrix, assigning positive scores for matching tokens and applying strict mathematical penalties for mismatches and the creation of "gaps"7. By tracing back the highest-scoring path, the algorithm generates an aligned variant graph. It strategically inserts virtual gaps into the text arrays where deletions or additions occur8. This ensures that identical paragraphs inherently align horizontally on the screen, even if preceding sections vary significantly in length or content.

4. Analysis and Variant Identification

The analysis module interprets the aligned variant graph to categorize the differences. It parses the relationships between the aligned texts and the virtual gaps to classify variants as additions, deletions, replacements, or transpositions8. However, traditional diff tools designed for software engineering inherently assume a chronological hierarchy (e.g., File A is the old version, File B is the new version), explicitly labeling text as "inserted" or "deleted." In an archival context, chronological precedence is often unknown or contested. Therefore, the workbench treats all panes equitably. It does not label text as definitively "inserted" or "deleted," but neutralizes the terminology, identifying all divergences simply as "variants."

Comparison Modes and Caveats Matrix

The workbench is fundamentally agnostic to the type of record being compared, provided it can be resolved to a readable text stream. The engine gracefully handles disparate input types, acknowledging the significant epistemological caveats of comparing exact-source PDFs (where physical formatting is locked) against extracted plain text.

The Exact-Source vs. Extracted-Text Caveat

A critical challenge arises when comparing an exact-source PDF against a manually transcribed Research Document. When processing an exact-source PDF, the workbench relies on the hidden text layer (whether OCR-generated or digital-born). Inevitably, the hard line breaks, hyphenations, and spatial layout of the PDF will not match the continuous prose of a plain text transcript. The Needleman-Wunsch engine mitigates this via the tokenization and normalization steps, actively ignoring visual line breaks to align the logical flow of words. However, to maintain absolute transparency, a specialized disclaimer icon (a classic yellow triangle warning) appears in the status bar. This warns the user that layout-level formatting, spatial mapping, and physical page geometries are completely excluded from the current comparison view.

Synchronization, Navigation, and Scroll Behavior

The mechanics of scrolling and navigating parallel documents define the ultimate usability of the workbench. Researchers must be able to seamlessly read two documents that differ wildly in length without constantly losing their spatial alignment.

Synchronized Scrolling and Proportional Mapping

Standard synchronized scrolling algorithms blindly move both panes by an equal number of vertical pixels. If Document A is 10 pages long and Document B is 50 pages long, equal-pixel scrolling rapidly desynchronizes the texts, rendering the tool useless. The workbench must implement Proportional Scrolling based on token alignment rather than raw pixel counts or simple document percentages. When the user scrolls the active pane, the application calculates the current visible token range in the viewport. It then interrogates the variant graph to map this range to the corresponding aligned tokens in the adjacent pane, automatically adjusting the scroll position of the secondary pane to ensure the matching text remains horizontally parallel31. If a massive structural gap exists (e.g., Document B contains a 10-page chapter entirely missing from Document A), proportional scrolling dynamically compensates. As the user scrolls through the unique 10-page chapter in Document B, the scrollbar of Document A remains temporarily locked at the precise locus of the variant. Once the user scrolls past the gap and reaches a shared alignment point, Document A immediately resumes synchronized scrolling.

Scrollbars as Progress Indicators

In classic desktop UI design, scrollbars are not merely interactive controls for movement; they serve as critical progress indicators representing the relative amount of the document visible in the window33. The workbench implements proportional scroll boxes (the draggable "thumb" of the scrollbar), which adjust their vertical size dynamically to reflect the percentage of the document currently loaded in the viewport33. This provides the user with immediate, tactile feedback regarding document length disparities.

Keyboard Navigation and Accessibility

To emulate historical MDI applications and ensure robust accessibility, full keyboard navigation is strictly required. The interface must not rely solely on mouse interactions.

• F6: Cycles focus between the split panes within a single MDI child window, moving the active cursor between Document A and Document B.
• Ctrl+F6: Activates the next MDI child window in the global Z-order, bringing a different comparison session to the foreground20.
• Ctrl+F4: Closes the currently active MDI child window20.
• Ctrl+Tab: Cycles through the tabbed property sheets located at the bottom of the window (e.g., switching rapidly between the Comparison View, Metadata View, and Concordance View).
• Alt+- (Minus): Opens the system menu of the active MDI child window, providing keyboard access to Move, Size, Minimize, and Maximize commands20.

The user interface must adhere to strict high-contrast color palettes typical of classic operating systems (e.g., Windows Standard or Windows Classic themes), ensuring that text remains legible and variant highlighting does not violate accessibility standards for color blindness.

Search, Find, and Cross-Document Concordance

While side-by-side reading handles linear textual flow, the workbench also provides macro-level statistical analysis via a dedicated Concordance Property Sheet accessible as a tabbed panel at the bottom of the MDI child window. This mirrors the functionality of powerful historic concordancers like TACT and AntConc1.

Independent vs. Linked Find

The traditional Ctrl+F Find dialog is enhanced to accommodate the parallel environment, governed by a classic modal property sheet.

• Independent Find: When unchecked, the search query applies only to the currently active pane. This is vital when a researcher is looking for a specific anchor word in one text to manually establish a baseline before analyzing the adjacent text.
• Linked Find: A checkbox on the Find dialog. When active, searching for a term executes a shared search-term highlighting across all open panes simultaneously. Activating the "Find Next" command jumps both panes to the nearest token match, prioritizing chronological alignment and leveraging the proportional scroll engine to bring both instances into the center of their respective viewports.

Keyword in Context (KWIC)

The KWIC view is a staple of digital humanities research. It strips away the traditional document hierarchy to present a list of search terms centered perfectly in a tabular list view control, surrounded by a user-defined [Figure omitted from source export] words of left and right context1. This display allows researchers to rapidly identify collocation patterns—how specific operational words or ideological terms are used differently across two compared documents. The UI allows users to sort the tabular data by the first word to the left (1L) or the first word to the right (1R), grouping similar semantic phrases alphabetically to reveal underlying syntactical habits10.

The Concordance Plot (Barcode Visualization)

Adapted directly from the visual paradigms introduced by AntConc, the Concordance Plot provides a high-level, macro-visual representation of term frequency and distribution1. Each document in the comparison session is represented as a horizontal rectangular track. Every occurrence of a search term is drawn as a thin vertical black line (a "barcode" hit) within the rectangle. This visualization is profoundly effective for structural analysis. It allows a researcher to instantly perceive if a specific geopolitical concept is heavily clustered in the introductory chapter of Document A, but scattered evenly throughout the entirety of Document B. It reveals the pacing and thematic structure of a document at a single glance1.

Term Frequency and Word Lists

A standard grid control displays total token counts, unique word counts, and comparative frequencies between the witnesses. If Document A uses a specific noun 45 times and Document B uses it only 2 times, the statistical anomaly is mathematically flagged. This offers immediate insight into shifting thematic focuses, vocabulary richness, or potential authorship changes without requiring the researcher to perform a full manual read of the texts11.

Methodological Safeguards and Source Provenance

Text-difference highlighting poses a significant epistemological risk within archival contexts. Standard software engineering diff tools inherently frame one document as the objective baseline and the other as a subjective deviation, alteration, or error2. This visual hierarchy creates a false interpretive authority, subtly instructing the researcher which text is the "original" and which is the "copy."

Mitigating False Interpretive Authority

To safeguard against this methodological trap, the workbench employs a strictly neutral "Derived Comparison" labeling system. Differences are not highlighted using aggressive red/green strike-throughs. Instead, variance is visualized using a subtle heatmap or grayscale intensity approach9. The background saturation of a text block deepens based on the extent of the variance, highlighting the locus of variation without establishing a base-text supremacy9.

Source-Provenance Messaging

To constantly reinforce the nature of the tool, the UI explicitly stamps the view with permanent provenance messaging. The status bar at the bottom of the MDI child window constantly displays the text: "Derived Comparison View – Not an Authoritative Source." Furthermore, hovering over any variant block summons a classic Windows tooltip displaying the exact Hash ID of the source document, ensuring the researcher is continually aware that they are looking at a mathematical alignment of tokens, not the physical artifact itself.

Workflows and Ecosystem Integration

The Parallel Reader is not an isolated utility; its value stems from deep integration with the broader Al.Qaeda.net archival ecosystem.

Drag and Drop Initialization

Researchers do not need to navigate complex, modal file selection dialogs to initiate a comparison. The workbench supports native OLE-style drag-and-drop operations from other modules. A user can open the Archive Explorer or their local Research Trail list, click and hold a record icon, and drag it directly into the empty gray workspace of an MDI child window. Dropping a second record into the window automatically triggers the Needs-Wunsch collation engine and splits the pane horizontally or vertically based on the user's default preferences.

Integration with the Cognitive Atlas

If a researcher is exploring the node-based Cognitive Atlas (e.g., viewing a specific organizational cell or operational codename), they can right-click the node to access a context menu. Selecting Compare Primary Witnesses automatically instantiates the Comparison Workbench, loads the top two foundational documents associated with that node, and immediately executes a Linked Find for the codename. This seamless workflow centers both panes on the relevant paragraphs, allowing the researcher to jump instantly from high-level conceptual mapping to granular textual criticism.

Local Privacy Architecture and Data Serialization

A rigid, non-negotiable requirement of the Al.Qaeda.net reconstruction is absolute data locality and privacy. Canonical source payloads are strictly protected, and any user-provided files, temporary documents, or comparison session data must never touch a remote server or be transmitted over the network.

Local-Only Saved Comparison Sessions

When a user executes a comparison, the Needleman-Wunsch algorithm generates a highly complex, memory-intensive variant graph. Recalculating this graph for large texts upon every viewing is computationally prohibitive. Therefore, the workbench allows users to utilize a standard "Save Session" command. This action serializes the state of the comparison directly into the browser's local IndexedDB or prompts the user to download a .cmp (Comparison) JSON wrapper file to their local hard drive. To ensure the canonical text payload is not illegally exported, bypassed, or duplicated, the session file stores only:

1. The unique Hash IDs and local URIs of the canonical sources.

2. The UI state (pane split ratio, window coordinates, toggle states, normalization preferences).

3. User-generated annotations, tied strictly to their positional anchors (e.g., Token index 450 to 455).

Upon loading a .cmp file via the File menu, the workbench parses the JSON, retrieves the required text sources from the local, encrypted archive cache using the Hash IDs, and reconstructs the variant graph and visual view dynamically. This ensures total data compliance.

Exporting Reports via TEI Standoff Markup

Researchers frequently require exports of their findings for publication or external review. However, exporting a document that contains both their notes and the copyrighted/protected source text violates payload protection constraints. To resolve this, the workbench utilizes a methodology inspired by the Text Encoding Initiative (TEI) P5 guidelines for "Standoff Markup"37. In standoff markup, analytical annotations and variant tags are stored in a completely separate file from the primary text data. The markup references the base text strictly via structural pointers (such as xml:id or token offsets)37. The workbench's Export feature generates a TEI-compliant XML or JSON report detailing where variations occur (e.g., "Document Hash X, Token 450 diverges from Document Hash Y, Token 432"). It includes all user-generated notes and concordance frequencies, but entirely omits the protected text strings. This provides the researcher with a rigorous, machine-readable, and highly citable research artifact without exporting the underlying canonical payload.

Performance Limits, Risks, and Implementation Strategy

While historical tools processed texts in milliseconds due to small file sizes and rudimentary algorithms, modern web environments handling full book-length comparisons using dynamic programming face significant performance bottlenecks.

Algorithmic Complexity and Long Texts

The Needleman-Wunsch algorithm has a time and space complexity of [Figure omitted from source export], where [Figure omitted from source export] and [Figure omitted from source export] are the token counts of the two documents. Comparing two 10,000-word chapters requires evaluating a matrix of 100,000,000 cells29. This will reliably freeze the main JavaScript thread, breaking the classic UI illusion and triggering browser unresponsiveness warnings. To mitigate this, the collation engine is strictly offloaded to Web Workers. When a large comparison is initiated, the UI remains responsive, and the Status Bar displays a classic Windows-style progress bar (segmented blue blocks) indicating the worker thread's status. For documents exceeding 15,000 words18, the engine implements heuristic chunking—breaking the text at reliable anchors (such as chapter headings, paragraph breaks, or prolonged identical strings) and aligning the chunks sequentially rather than calculating a single massive matrix.

Misuse and Confusion Risks

Deploying advanced textual analysis tools to general researchers carries inherent risks of misinterpretation.

1. The OCR Fallacy: Users may compare a degraded historical scan against a modern clean transcription and interpret the vast sea of highlighted variants as a conspiracy of deliberate alteration. In reality, these are often just OCR artifacts. The engine defaults to normalizing common OCR errors and displays a tooltip warning when comparing derived text against digital-born text.

2. The Illusion of Completeness: A proportional scrollbar can make two documents look structurally identical until the user realizes Document B is missing an entire appendix. The Concordance Plot mitigates this by visualizing structural length disparities clearly; if Document B ends prematurely, the plot shows a massive empty space, visually communicating the truncation.

3. MDI Resource Exhaustion: Users might attempt to open 15 MDI child windows simultaneously, exhausting browser memory and cluttering the DOM. A hard limit of 8 active MDI child windows is enforced. Attempting to open a 9th window triggers a classic Win32 error sound and a modal message box.

Phased Implementation Priorities

To ensure stable delivery and manage technical complexity, the workbench will be implemented in three distinct phases.

Concrete Acceptance Tests

To validate the integrity, performance, and historical authenticity of the feature, the following rigorous acceptance criteria must be successfully executed prior to deployment.

1. MDI Confinement and Z-Order Test:

• Action: Drag an inner child window (Comparison Session) toward the edge of the screen. Open three windows and click them in random order.
• Expected Result: The child window must be clipped by the bounds of the parent MDI client area; it cannot float over the OS desktop. Clicking a window brings it immediately to the front of the Z-order without UI tearing or lag.

2. Proportional Synchronization Test:

• Action: Load a 100-word document in Pane A and a 5,000-word document in Pane B. Scroll Pane B rapidly from top to bottom.
• Expected Result: Pane A remains anchored to the closest corresponding token, remaining completely stationary until the specifically aligned section of Pane B enters the viewport, at which point it resumes scrolling.

3. Local Privacy and Air-Gap Test:

• Action: Load a local text file, generate a comparison graph, and save the session. Physically disconnect the internet connection. Reload the application and load the session file.
• Expected Result: The session restores perfectly without requiring network calls. Network monitoring confirms exactly zero bytes were transmitted during the collation process.

4. Standoff Export Verification Test:

• Action: Add a custom user note to a variant block. Export the report to JSON/XML via the Export Wizard.
• Expected Result: The resulting file contains the user note, the token coordinates, and the Canonical Hash IDs, but contains exactly zero words of the protected source text payload, passing all copyright checks.

5. Keyboard Accessibility Test:

• Action: Press Ctrl+F6 while three comparison windows are open, then press F6 to swap panes.
• Expected Result: Focus cycles through the child windows according to standard Z-order rules, updating the active title bar colors from inactive gray to active dark blue. F6 smoothly transfers cursor focus between the left and right comparison panes.

By adhering strictly to this architectural blueprint, the Parallel Reader and Concordance Workbench will provide a methodologically sound, mathematically rigorous, privacy-respecting, and period-authentic instrument. It will vastly expand the analytical capabilities of the archival reconstruction, allowing researchers to uncover the silent evolution of historical texts.

Works cited

1. AntConc, historians and their diverging research methods, https://dhh.uni.lu/2020/08/11/antconc-historians-and-their-diverging-research-methods/

2. Collation in Scholarly Editing: An Introduction, https://wraabe.wordpress.com/2008/07/26/collation-in-scholarly-editing-an-introduction-draft/

3. SQLskills Whitepaper: SQL Server DBA Guide to the SQLCLR, https://www.sqlskills.com/resources/whitepapers/sql%20server%20dba%20guide%20to%20sqlclr.htm

4. Kenneth M. Price and Ray Siemens, Introduction, https://dlsanthology.mla.hcommons.org/introduction/

5. Using TACT with Electronic Texts: A Guide to Text Analysis, https://search.proquest.com/openview/3b7d0a1be2518eb2c3d531b05bd891a1/1?pq-origsite=gscholar\&cbl=48796

6. Discover research tools for studying texts. \- TAPoR, https://tapor.ca/tools?tag\_values=624

7. Documentation \- CollateX, https://collatex.net/doc/

8. Detecting Transpositions when Comparing Text Versions using, https://dragonfly.hypotheses.org/954

9. A Survey of Text Alignment Visualization \- IEEE Computer Society, https://www.computer.org/csdl/journal/tg/2021/02/09220137/1nRLN0fSuiI

10. kwaldenphd/AntConc-tutorial \- GitHub, https://github.com/kwaldenphd/AntConc-tutorial

11. (PDF) AntConc : A Learner and Classroom Friendly, Multi-Platform, https://www.researchgate.net/publication/33010037\_AntConc\_A\_Learner\_and\_Classroom\_Friendly\_Multi-Platform\_Corpus\_Analysis\_Toolkit

12. Developing a Freeware, Multiplatform Corpus Analysis Toolkit for, https://laurenceanthony.net/research/tpc\_2006\_anthony\_antconc\_paper.pdf

13. Concordancer \- Wikipedia, https://en.wikipedia.org/wiki/Concordancer

14. Wordsmith Tools as an Enabler for Text Analysis \- UP Journals, https://upjournals.up.ac.za/index.php/dhasa/article/download/3824/3529/21223

15. You're Collating Just Fine and Other Lies You've Been Telling Yourself, https://journals.openedition.org/digitalmedievalist/2482

16. Text Tools \- A Companion to Digital Humanities, https://companions.digitalhumanities.org/DH/?chapter=content/9781405103213\_chapter\_33.html

17. Computer-supported collation of modern manuscripts: CollateX and, https://pure.knaw.nl/ws/files/797964/BDMP\_Collatex\_LLC\_article\_2013\_05\_24\_submitted\_20130529.pdf

18. Computer-supported collation of modern manuscripts: CollateX and, https://www.researchgate.net/publication/270752678\_Computer-supported\_collation\_of\_modern\_manuscripts\_CollateX\_and\_the\_Beckett\_Digital\_Manuscript\_Project

19. Multiple Document Interface \- Win32 apps | Microsoft Learn, https://learn.microsoft.com/en-us/windows/win32/winmsg/multiple-document-interface

20. About the Multiple Document Interface \- Win32 apps | Microsoft Learn, https://learn.microsoft.com/en-us/windows/win32/winmsg/about-the-multiple-document-interface

21. 19-1 oop.indd \- revista Acta Universitaria \- Universidad de Guanajuato, http://www.actauniversitaria.ugto.mx/index.php/acta/article/viewFile/120/106

22. Multiple Document Interface (MDI) \- Lazarus Forum \- Free Pascal, https://forum.lazarus.freepascal.org/index.php?topic=45486.0

23. Standard MDI \- Actipro WinForms Controls Docs, https://www.actiprosoftware.com/docs/controls/winforms/docking/standard-mdi

24. Method \- Frankenstein, https://frankensteinvariorum.org/method

25. The Gothenburg model and CollateX, https://automaticcollationlausanne2020.github.io/session4a.html

26. Juxta Web Service, LERA, and Variance Viewer. Web based ... \- RIDE, https://ride.i-d-e.de/issues/issue-11/web-based-collation-tools/

27. Text Alignment in the Service of Text Reuse Detection \- MDPI, https://www.mdpi.com/2076-3417/15/6/3395

28. Measuring similarities and visualizing patterns in a text tradition, https://academic.oup.com/dsh/advance-article/doi/10.1093/llc/fqag068/8698901

29. RNAcode\_Web – Convenient identification of evolutionary ... \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC10757073/

30. Design and Implementation of Architectures for Interactive Textual, http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-56601

31. Application programming interfaces for scrolling operations, https://patents.google.com/patent/US7844915B2/en

32. Portable electronic device, method, and graphical user interface for, https://patents.google.com/patent/US8223134B1/en

33. Apple human interface guidelines \- Andy Matuschak, https://andymatuschak.org/files/papers/Apple%20Human%20Interface%20Guidelines%201987.pdf

34. Microfeatures I love in blogs and personal websites | Hacker News, https://news.ycombinator.com/item?id=40774277

35. User Agent Accessibility Guidelines 1.0 \- W3C, https://www.w3.org/WAI/UA/WD-UAAG-20000115/

36. Kirstin Fierro \- The Incubator, https://incubator.domains.trincoll.edu/wp-content/uploads/2019/03/Kirstin-Fierro-Final-Paper-0031.pdf

37. Balisage Paper: Why TEI stand-off annotation doesn't quite work, https://www.balisage.net/Proceedings/vol5/html/Banski01/BalisageVol5-Banski01.html

38. 17 Linking, Segmentation, and Alignment \- The TEI Guidelines, https://www.tei-c.org/release/doc/tei-p5-doc/en/html/SA.html

39. issues, solutions and new trends in scholarly publishing within the, https://amsdottorato.unibo.it/id/eprint/4766/1/peroni\_silvio\_tesi.pdf