Dialog pictograms

In Dofus Unity, there are dialog pictograms that foretold the outcome of dialogs. For example, showing quest progression, or showing a quest is aviable. But actually, they weren't there for twenty years ! They've been added automatically on replies with given actions. However, previous dialogs have not been written taking this system into account, and sometimes it can be weird, misleading, or vanilla ruin a dialog.

I received an Excel extract of 18 000+ lines of pictogram appearances, and I had to find as many bad pictograms cases as possible and solve them.

Step 1 : Understand

Before anything else, I had to understand the Excel table, and make sense from it. I started doing early game quests, writting down information, and making links the data. It is important to understand that the extract was generated from replies (located in NpCs), while the user experience is driven by quests. That means the data order doesn't make much sense while testing.

These early tests were also the opportunity to witness dialog structures and learn how NpCs work.

Step 2 : Categorize

During my tests, I encountered several bad cases, and some were really similar. I hence used these cases to make a categorisation of bad cases, that will be used for an automated mass detection.

I came up with 4 categories :

Shortcuts

Sometimes, there are several ways to exit a dialog, and sometimes, you see an exit before the other. This wasn't a problem before, but now there's a "?" on that exit, so players will likely click it and miss a part of the dialog, or at least it will alterate their choice.

The easiest solution found for these cases is to move one of the feadbacks. However, this may require some rewriting.

Systematic check

Sometimes, we have to check an objective at the end of a dialogue even though there are several exit replies. This is especially common for "talk to" objecives : Say what you want, at the end, you've talked. In order to do that, there was typicially a validation on each exit reply, but now, they show a question mark, misleading player into thinking there's a choice, whereas there is not.

The easiest solution found for these cases is to merge both validations before the fork. However, this may require some rewriting.

Double Pictos ?X

Sometimes, there's an objective validation and a fight trigger on the same reply. It generates two pictos, which is a little weird, and may be confusing. Most of the time, the "?" is actually a "talk to" validation, but is not interpreted like that. It actually is interpreted as a "fight that makes my quest progress".

This is also bad for the game's consistency since there are fight triggers without validations. Also, if you lose the fight and come back, the "talk to" objective being already validated, there will be only swords, which looks like a bug.

Triple Pictos ??X

Sometimes, you can start a fight, but also say no, and fight later. But now there's a "?" on the flee option, making it looklike a real option, whereas no, it's just validating the "talk to" objective ! And if you flee and come back, the "talk to" objective being validated, both "?" will be missing, once again looking like a bug like a bug.

This case is absolutely the most misleading and I decided to hunt it down.

To solve these cases, we can either uncouple the fight trigger and the validations, and merge them like we've done in the systematic check case, but we can also hide the "?" because it goes very smoothly without them.

Step 3 : Automatize

With my categorization, I was able to write a script that would find cases across the whole game. I learned how to use excelscript for the occasion and made a search tool, usable directly from the sheet.

The script has two steps :

Gather table lines with something in common. For example, same reply_id, meaning one reply has several actions.
Thining based on criterions in order to sharpen the selection.

I've made two presets :

?X+ : Finds replies that have both an objective validation and a fight trigger. It doen't see the flee option, but when we look at the NpC's structure, we see it.

This filter has 100% efficiency (all candidates are actual cases, and no known case is ignored by the filter)

Shortcuts and ?? : Finds parameters used multiple times (meaning several actions do the same thing), with an objective validation and different texts and same NpC and same dialog and same reply display criterion

This filter has 25% efficiency (Most candidates are not actual cases, but it still finds a lot of cases)

Conclusion

This work on pictograms was definetly my own big undertaking. People knew there were going to be issues somewhere, but no one knew what et where exactly. I really enjoyed taking it into my hands and structuring it.

It was also my first time working with this much data and I'm proud of my tool able to filter the whole table within four seconds. I'm also very proud that my tool is documented and accessible to anyone, so the one who will continue this mammoth project after me will not start empty handed.