BKK FIRMUS
|| Handwritten text recognition

↑

To analyze text comprehension from given cases, we built a minor study with 28 observations from diverse participants. The objective of this study is to understand how a human could interpret letters and numbers without prior training in terms of expected values in a certain context. The collection of answers are displayed bellow, together with our models’ and ChatGPT5 responses in each case.

Code

hand = read_excel("00_data/Handwritting comprehension (respuestas).xlsx",
             sheet = "resp",
             col_names = TRUE,
             na = "",
             trim_ws = TRUE) %>%
    clean_names() %>% 
  filter(!grepl("chatgpt|correct",marca_temporal))

Case a

Correct answer: Max Mustermann Knollstr.16
Sadnacaya Model: Max Mustermann Krollstr.16
ChatGPT5: Max Mustermann Knollstr.16

Case b

Correct answer: 051 | 388715 Danke!
Sadnacaya Model: 0511 388715 Danke !
ChatGPT5: 0511 388715 Danke !

Case c

Correct answer: O645 483167.
Sadnacaya Model: 0645 483167.
*ChatGPT5: 0645 483167.

Case d

Correct answer: Ledammer Bachstraße
Sadnacaya Model: Gedammer Bachstraße
ChatGPT5: Ledammer Bachstraße

Case e

Correct answer: ewig krank
Sadnacaya Model: ewig krank
ChatGPT5: ewig brennt

General Observations