Updated configuration for html text translation to work in wasm test page

[abhi-agg]

Added alignment: soft during translator construction as per #266 (comment) and ResponseOptions::alignment to true as per #266 (comment)

Both of these changes were required for the html text translation to work. Only providingalignment: soft during translator construction and not setting ResponseOptions::alignment to true doesn't give correct results.

[jerinphilip]

Can you please post a few screenshots confirming this works? I'm looking for examples where @jelmervdl's length based approximations are not giving a false pretence of a working HTML translation.

I'd expect ResponseOptions.alignment to be required to set to true for buildAlignments to be run. HTML.Restore I expect consumes the product of buildAlignments.

[jerinphilip]

Hello, an update. ResponseOptions.alignment will need to be set for the time being. I had an issue with my python package (loading from a venv instead of my source).

With options incorrect

Source

[src]  In early 2018, security researchers disclosed two major vulnerabilities, known as <a href="https://en.wikipedia.org/wiki/Meltdown_(security_vulnerability)" target="_blank">Meltdown</a> and <a href="https://en.wikipedia.org/wiki/Spectre_(security_vulnerability)" target="_blank">Spectre</a>.
[hyp]  Anfang 2018 enthüllten Sicherheitsforscher zwei große Schwachstellen, bekannt als Meltdown <a href="https://en.wikipedia.org/wiki/Meltdown_(security_vulnerability)" target="_blank">und Spec</a>tre<a href="https://en.wikipedia.org/wiki/Spectre_(security_vulnerability)" target="_blank">.</a>


[src] >  <p>I <b>like</b> to <i>drive</i> this car.</p>
[hyp] >  <p>Ich <b></b>fahrer<i>e dieses</i> Auto gerne.</p>

Render

[src] In early 2018, security researchers disclosed two major vulnerabilities, known as Meltdown and Spectre.
[hyp] Anfang 2018 enthüllten Sicherheitsforscher zwei große Schwachstellen, bekannt als Meltdown und Spectre.

[src]

I like to drive this car.

[hyp]

Ich fahrere dieses Auto gerne.

With correct options:

Source

[src] In early 2018, security researchers disclosed two major vulnerabilities, known as <a href="https://en.wikipedia.org/wiki/Meltdown_(security_vulnerability)" target="_blank">Meltdown</a> and <a href="https://en.wikipedia.org/wiki/Spectre_(security_vulnerability)" target="_blank">Spectre</a>.
[hyp] Anfang 2018 enthüllten Sicherheitsforscher zwei große Schwachstellen, bekannt als <a href="ttps://en.wikipedia.org/wiki/Meltdown_(security_vulnerability)" target="_blank">Meltdown</a> und <a href="https://en.wikipedia.org/wiki/Spectre_(security_vulnerability)" target="_blank">Spectre.</a>


[src]   <p>I <b>like</b> to <i>drive</i> this car.</p>
[hyp]  <p>Ich <i>fahrere</i> dieses Auto gerne.</p>

Render

[src] In early 2018, security researchers disclosed two major vulnerabilities, known as Meltdown and Spectre.
[hyp] Anfang 2018 enthüllten Sicherheitsforscher zwei große Schwachstellen, bekannt als Meltdown und Spectre.

[src]

I like to drive this car.

[hyp]

Ich fahrere dieses Auto gerne.

[abhi-agg]

I have attached Wasm test page screenshot that translates the highlighted text from the wikipedia webpage. It is a bit difficult to read everything in the wasm test page. So I copied both source and translation in https://wordtohtml.net/ that gives nicely formatted html and attached the screenshot for that as well.

The highlighted text from the wikipedia webpage:

Wasm test page:

wordtohtml:

[kpu]

I'm confused as to why we're making Abhishek bend around configuring things. If HTML is set we turn alignments on.

[abhi-agg]

Is there a performance penalty if alignment: soft is always provided (during translator construction as it is right now) but user never translates html text but only plain text?

If not, then alignment: soft could become the default, relieving user from specifying it explicitly during translator construction. This way, whenever user provides ResponseOptions::HTML as true during translate call, the html text translation kicks in (I think that's what Kenneth means above?). Otherwise plain text translation kicks in.

[kpu]

There is a speed penalty having alignment: soft (as opposed to one). Could also be a RAM penalty but we're loading the full file anyway. I'm not sure how big it is. For firefox it may as well always be on though. We should always calculate alignments if HTML is on and you shouldn't need to configure both at the request level.

[abhi-agg]

Thanks @kpu for the explanation. I am going ahead with merging this one to have the main branch working for html text translation. I will change config again based on the outcome of #270 👍🏾