On the eve of the publication of the NIOD report on Srebrenica, a retrospective of the Excess Memorandum that was presented in June 1969. Commissioned by the government, archive research was conducted into the war crimes committed by Dutch soldiers in Indonesia in the years 1946-1950. The immediate reason for the investigation was an interview in the current affairs section "Achter het Nieuws" with Dr. J. Hueting. The interview caused an enormous commotion and brought to the surface the national trauma of the colonial war.
Automatic Translation, does it work?
Andere Tijden is a program in the Netherlands about past events: the second world war, sport events, missions abroad, political issues in the past etc. Some of the broadcast may be important for non-Dutch but a full, time aligned translation may be to much work (and cost).
So, we want to figure out if Automatic Translation (AT) may be of use. Of course, AT, like ASR, makes mistakes, but is the chain ASR -> Phrase Generation -> AT at least good enough to give listeners/viewers a clear idea about what the AV-recordings are about? Phrase Generation is the proces of making "phrases" from the string of adjacent recognised words. Speech recognition is producing recognised words but all punctuation marks are absent. AT needs these phrases, including the punctuation marks to produce a reliable translation.
In the recording showed below, we have used ASR for the transcript, made phrases by hand without editing the transcripts, and then used Google Translate to produce the English subtitles.
The first tab shows the possiblities of ASR for Dutch and automatic translation into English with Google Translate. The second tab shows the recognition with a "Karaoke-style" view of the results (Dutch only).