Salvaging scraps of conversation from the wasteland.

Ever heard of Van Buren? That's the original (cancelled) first Fallout 3 game codename.

The motivation behind this script is to generate a dataset that I can use to experiment some computational linguistics and NLP methods on a corpus that captures the grit, slang, and broken etiquette of the Fallout speech. With that linguistic rubble in hand, you can play with topic models to unearth recurring rumors, build language models that mimic irradiated idioms or just train classifiers and measure how some patterns shape in-game dialects.

This Pascal/Delphi unit is a FNVEdit (xEdit FNV variant) script that scans loaded plugin files and exports select dialogue-related records and related metadata to a CSV file.

• Recursively scans GRUP containers to export INFO, MESG, DIAL, NOTE, and LSCR records to a CSV.
• Extracts text (NAM1/DESC/FULL/TNAM), menu button ITXT, content source, file name, FormID, EditorID, name, topic, quest, top-level path, and speaker.
• Multiple response lines or menu entries are joined with |.
• Encodes CR/LF as \n, quoting fields safely (doubling quotes and wrapping when needed).
• Reports progress every 500 records with a completion summary.

• Fallout New Vegas game (with or without extensions) installed on your computer.

• Properly configured FNVEdit (or there). Tested with version 4.5.1

1. Place the unit into your xEdit/script folder.
2. Load plugins in the appropriate xEdit build (FNVEdit, etc.) and select the appropriate .esm (or others) files.
3. Right click and select apply script on the left pane:
   • 
   • then choose and run FNV_dialogue_export from the list:
   • 

• Output file: <wbTempPath>Fallout_New_Vegas_dataset.csv
  • Usuallly located within: C:/Users/YourWindowsUserName/AppData/Local/Temp, where YourWindowsUserName is your actual user name.

• CSV header:

signature;file_name;formID;editorID;text_content;content_source;name;topic;quest;topLevelPath;additional_content;speaker

• Output location: the CSV is written to wbTempPath + 'Fallout_New_Vegas_dataset.csv'. Modify the OutFile assignment in Initialize per your needs.
• CSV delimiter: to switch separators (for example to comma), replace each ';' concatenation in ScanContainer and the header string in Initialize with the desired delimiter.
• Recorded signatures: to include additional record types, add their signature checks in ScanContainer (e.g., SigEquals(sig, 'SIGN')) and implement an extraction logic for their fields.
• Progress interval: change the frequency of progress messages by editing the condition (RecordCounter mod 500 = 0) to another modulus value.
• Locale/encoding: TStringList.SaveToFile uses system default encoding. For UTF-8 output, use SaveToFile(OutFile, TEncoding.UTF8) or an equivalent streaming write.

Top-level file structure
PLUGIN (.EXE / .ESM / .ESP)
│
├── GRUP Top "DIAL"
│   └── DIAL (Dialogue Topic)
│       └── GRUP Topic Children
│           └── INFO (Dialogue Response)
│
├── GRUP Top "MESG"
│   └── MESG (Message)
│
├── GRUP Top "NOTE"
│   └── NOTE (Note / Terminal Entry)
│
├── GRUP Top "LSCR"
    └── LSCR (Loading Screen)

DIAL — Dialogue Topic

DIAL
│
├── EDID                → EditorID
├── FULL - Name         → Topic text (player-visible)
│
├── DATA - DATA
│   └── Type            → Topic type
│
├── Record Header
│   └── FormID
│
└── GRUP Topic Children
    └── INFO [...]

INFO — Dialogue Response (most complex)

INFO
│
├── EDID
├── ANAM - Speaker      → Speaker (NPC/reference)
│
├── Topic               → Parent topic reference
│
├── DATA - DATA
│   └── Type            → INFO subtype
│
├── QSTI - Quest        → Owning quest (if any)
│
├── Responses
│   ├── Response
│   │   └── NAM1 - Response Text
│   ├── Response
│   │   └── NAM1 - Response Text
│   └── ...
│
└── (implicit parent path via DIAL)

MESG — Message Box

MESG
│
├── EDID
├── FULL - Name         → Message title
├── DESC - Description  → Main message text
│
├── Menu Buttons
│   ├── Menu Button
│   │   └── ITXT - Button Text
│   ├── Menu Button
│   │   └── ITXT - Button Text
│   └── ...

NOTE — Notes / Terminal Notes

NOTE
│
├── EDID
├── FULL - Name
│
├── DATA - Type         → Note type
│
├── TNAM - Text / Topic
│   └── Text            → Note content
│
├── Quests
│   └── ONAM - Quest    → Linked quest

LSCR — Loading Screen

LSCR
│
├── EDID
├── DESC - Description → Loading screen text

• Environment dependent: does not run as a standalone Delphi program. Requires xEdit runtime functions (ElementByName, GetEditValue, FileCount, wbTempPath, AddMessage, etc.) that are accessible within FNVEdit.
• Record coverage: only scans INFO, MESG, DIAL, NOTE, and LSCR records (and GRUP containers). Other records are ignored unless implemented.
• Performance the ScanContainer procedure is recursive and will traverse GRUP/group hierarchies by calling itself for each nested group. This approach is slower in terms of performance.
• CSV separator: Uses ; as the delimiter. If downstream tools expect ,, import errors or incorrect parsing may occur.
• Field assumptions: Extracted fields are based on common paths used in Bethesda/FO editors; custom or unexpected record layouts may yield empty or incomplete values.
• Memory usage: Loads all output lines into memory via TStringList before saving; very large exports may consume significant RAM.
• Progress granularity: Progress messages are emitted every 500 records.
• Error handling: SafeEditValue suppresses exceptions and returns empty strings; this hides field-specific errors but may mask data issues.

• Needs further adaptation to extract contents from other Bethesda games (Skyrim, Fallout 4..).

• Improve performance (limit the recursion, stream output from TStringList).

Feel free to open issues or discuss any improvements. Contributions are welcome! Each contribution and feedback helps improve this project and my skills - it's always an honour :)

Fork the repository.
Create a branch for your feature or bug fix.
Submit a pull request with your changes.

The authors of the xEdit library and especially the maintainers of the FNVEdit build.

Provided under the Unlicense license.

🀅