What does the (distant) future of language acquisition look like?

General discussion about learning languages
User avatar
Le Baron
Green Belt
Posts: 254
Joined: Mon Jan 18, 2021 5:14 pm
Location: Pays-Bas
Languages: English (N), Nederlands, Français, Deutsch, Esperanto (a very worthy language). Studying: Castellano, Swahili, rather slowly, but surely. Also Sranantongo in the past with my wife, but it has lapsed.
x 504

Re: What does the (distant) future of language acquisition look like?

Postby Le Baron » Mon Apr 12, 2021 3:23 pm

Iversen wrote:So when one of its problems is that it has holes in its vocabulary the correct way to cure it would be to feed it a dictionary or two and tell it to treat the translations in the dictionary as translations everywhere else.

The problem with vocabulary is GT can't understand either context or shades of meaning between words. In one of the better languages, French which I've played with in combination with several languages, it can't always differentiate the contextual usage (and/or similarity) of simple things like toujours/encore, car/parce que. However it does recognise lots of fixed phrases, which is no doubt useful. Feeding it vocab doesn't seem to eliminate the problem of nuance. Especially in grammar/syntax, which you duly noted.
0 x

User avatar
Iversen
Black Belt - 3rd Dan
Posts: 3212
Joined: Sun Jul 19, 2015 7:36 pm
Location: Denmark
Languages: Monolingual travels in Danish, English, German, Dutch, Swedish, French, Portuguese, Spanish, Catalan, Italian, Romanian and (part time) Esperanto
Ahem, not yet: Norwegian, Afrikaans, Platt, Scots, Russian, Serbian, Bulgarian, Albanian, Greek, Latin, Irish, Indonesian and a few more...
Language Log: viewtopic.php?f=15&t=1027
x 7937

Re: What does the (distant) future of language acquisition look like?

Postby Iversen » Mon Apr 12, 2021 4:00 pm

First step is to tell GT as many words with as many of their possible translations as possible. I use the program to make bilingual study printouts so I see a lot of translations from (and to) a wide array of languages, and most problems seem to be caused by a lack of vocabulary - or in a weaker form: lack of knowledge about certain meanings, which any decent dictionary could have told it. But often a certain word is totally missing from GT's vocabulary, and that can happen even with words you would have expected it to know.

The third step (according to my previous rant) is then to ask the machine to choose the most relevant translation in a given context, and there the systematic analysis of contexts plays a role - but more subtle shades of meaning will be hard to communicate to a machine that doesn't have feelings and doesn't have a life outside the virtual world.

However the point is not whether it can 'feel' the difference between for instance "pour que" and "parce que" as long as it can choose the correct translations in a given target language. And since there is a lot of difference between an intention and a reason most languages have some way of indicating which one is meant - GT just has to identify the different possibilities and choose the most logical one, given the context - like for instance whether there is a subjunctive in the subordinate phrase (and there the markers I mentioned might come in handy).

The need for some deep understanding of the semantics only arises at stage four, where a machine is told to construct new utterances.
1 x

User avatar
EGP
White Belt
Posts: 30
Joined: Sun Mar 28, 2021 8:36 pm
Location: Australia
Languages: English (N), Macedonian (B2), German (A1)
x 42
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby EGP » Mon Apr 12, 2021 8:10 pm

Iversen,

I wonder if you have heard of the Appen group? One of the jobs they keep sending me is for audio transcriptions, and others are for similar types of translation work often for this AI stuff. (I'll just briefly mention they also have Alexia and Facebook survey stuff I know less about).

I bring this up because, at the end of the day, it does take money to pay someone who is an expert in these fields to work with developing the software. And the more they feed real people to verify things, the better it gets.

Labour intensity is not the only issue. There are also environmental and ethical issues that creep in with AI. Somewhere in the world, there are computers burning tons of resources that could otherwise be used to feed those without access to essential ones. Then there is those in power growing more power and spreading the types of discourse that they will want the AI to express.

---

I think of apps like Grammarly that I use primarily because I have bad eyesight (I can't see missing letters or ones in the wrong order), and I type too fast for my own good. Grammarly corrects things at about 70% accuracy in my opinion. Language is so complex that even with their huge budget (I imagine) that's as good as it gets.

The grammar app that I am working on has all the same issues as mentioned above. 1. It costs me money to host it, it burns resources when it works, it costs me huge amounts of time to keep developing and testing, it still has trouble with all the data that it is based on and that I keep feeding it.

As you say, the meaning of 'ceiling' depends on context, but you'd have to write a code that traverses sentence boundaries looking for words that relate to one or the other of the senses of the word. That would burn huge amounts of resources.

What about the other 60,000 words and all the other languages to translate them into with another 60,000?

Endless amounts of data and time.
1 x
I research English grammar and vocabulary in corpora.

Cainntear
Black Belt - 1st Dan
Posts: 1930
Joined: Thu Jul 30, 2015 11:04 am
Location: Scotland
Languages: English(N)
Advanced: French,Spanish, Scottish Gaelic
Intermediate: Italian, Catalan, Corsican
Basic: Welsh
Dabbling: Polish, Russian etc
x 4945
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby Cainntear » Mon Apr 12, 2021 9:22 pm

Le Baron wrote:'Yours sincerely' or 'sincerely' for people you know or are familiar with. 'Yours faithfully' when you're addressing someone for the first time/you don't really know them/you're writing a one-off formal letter.

Never ever in an email. Except to an English school. English teachers are the only people who use "yours [whatever]" in email, in my experience.

It was always "Regards," in my workplace.
0 x

User avatar
Iversen
Black Belt - 3rd Dan
Posts: 3212
Joined: Sun Jul 19, 2015 7:36 pm
Location: Denmark
Languages: Monolingual travels in Danish, English, German, Dutch, Swedish, French, Portuguese, Spanish, Catalan, Italian, Romanian and (part time) Esperanto
Ahem, not yet: Norwegian, Afrikaans, Platt, Scots, Russian, Serbian, Bulgarian, Albanian, Greek, Latin, Irish, Indonesian and a few more...
Language Log: viewtopic.php?f=15&t=1027
x 7937

Re: What does the (distant) future of language acquisition look like?

Postby Iversen » Mon Apr 12, 2021 10:24 pm

EGP wrote:Iversen, I wonder if you have heard of the Appen group?


ahem, no..

EGP wrote:As you say, the meaning of 'ceiling' depends on context, but you'd have to write a code that traverses sentence boundaries looking for words that relate to one or the other of the senses of the word. That would burn huge amounts of resources.
What about the other 60,000 words and all the other languages to translate them into with another 60,000?


I could imagine translate software that built a 'semantic bubble' as it went through a text, and if it had met a lot of masonry-related OR economic/environmental vocabulary along the way it would know how to translate 'ceiling'. OK, it might be confused if the context was an article about upper limits in the house building industry - but it would still be better equipped to make a sensible choice than if it hadn't collected information along the way.

As for the huge numbers of words in each language: that's not a show-stopper. Most translation problems are tied to the most common (and irregular and ambiguous) words, so having 60.000 or more words per language in the system per language should just mean better translations, not more computing time. Besides Google Translate's translations pass through English (at least in the overwhelming number of cases, and that's also what my own tests have shown) so it doesn't even try to find direct translations between all the possible combinations of languages. This does have an impact on the quality of the translations, but so be it - even with the massive computing power available to Google the number of all possible combinations between 60.000 words in hundreds of languages would be too high, and the proces of collecting the information would probably be impossible due to a limited supply of bilingual texts between the smaller languages.

But maybe some kind of 'intelligent' control check could warn the system of typical errors caused by the use of the English language as an intermediary vehicle, and then it would of course be relevant to let the system search for direct connections or connections through other languages in specific cases. With enough computer power that shouldn't be impossible.
0 x

lichtrausch
Orange Belt
Posts: 241
Joined: Thu Jul 23, 2015 3:21 pm
Languages: English, Japanese, German
Learning: Mandarin, Korean, French
x 509

Re: What does the (distant) future of language acquisition look like?

Postby lichtrausch » Tue Apr 13, 2021 1:05 am

EGP wrote:I have trouble believing that AI can get near human-like communication. One reason is that there are just too many senses of language structures. Another is that the machine does not have a human body. As in our body is an extension of our senses. We communicate with more than words. We also have things like satire. We use persuasion for reasons etc.

Multimodal AI will help with this.

The immense potential and challenges of multimodal AI
Unlike most AI systems, humans understand the meaning of text, videos, audio, and images together in context. For example, given text and an image that seem innocuous when considered apart (e.g., “Look how many people love you” and a picture of a barren desert), people recognize that these elements take on potentially hurtful connotations when they’re paired or juxtaposed.

While systems capable of making these multimodal inferences remain beyond reach, there’s been progress. New research over the past year has advanced the state-of-the-art in multimodal learning, particularly in the subfield of visual question answering (VQA), a computer vision task where a system is given a text-based question about an image and must infer the answer. As it turns out, multimodal learning can carry complementary information or trends, which often only become evident when they’re all included in the learning process. And this holds promise for applications from captioning to translating comic books into different languages. [...]

OpenAI is reportedly developing a multimodal system trained on images, text, and other data using massive computational resources the company’s leadership believes is the most promising path toward AGI, or AI that can learn any task a human can. And in a conversation with VentureBeat in January, Google AI chief Jeff Dean predicted progress in multimodal systems in the years ahead. The advancement of multimodal systems could lead to a number of benefits for image recognition and language models, he said, including more robust inference from models receiving input from more than a single medium.
1 x

User avatar
EGP
White Belt
Posts: 30
Joined: Sun Mar 28, 2021 8:36 pm
Location: Australia
Languages: English (N), Macedonian (B2), German (A1)
x 42
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby EGP » Tue Apr 13, 2021 8:26 am

I like the 'semantic bubble' but I have no idea if Google runs a full-text analysis of the text prior to the translation. I guess not though. I also get that not all parts of the text will require a full-text analysis. Still depending on the length of the text, I can certainly say you will encounter words with more than one sense and the full-text analysis might as well just happen from the start.

I'd imagine:

1. first

word count of repeating phrases to decipher the main topics which usually would only need to be the nouns.

2. Then do the word and grammar swapping that check back to those main topics.

Still, when you have a short text as you mentioned, words are ambiguous since there is nothing to refer back to or forward to.

(After spending time with the thousand++ grammar points in the English Grammar Profile I still have many [I guess 20%] that I can't locate with text alone.
They require manual interpretation that form just can't help narrow down.)

---

Lichtraus I think that a machine can be given temperature sensors, cameras, audio receptors etc. and similar to the last point, it gets a greater context. For example,

A woman is standing in front of the mirror and asks if she looks 'hot'.

Depending on the temperature, image banks can be scanned and a good guess could be made. Usually, a sociopolitically correct one or a flattering remark should be made to appease tension that may arise "mirror mirror who is the fairest...."

But the machine doesn't have the body, the human drive to reproduce, etc. I'm sure dolls could be programmed for that industry ... Oh hell I have no idea where my mind is going. 'blade runner'

dunno.
0 x
I research English grammar and vocabulary in corpora.

User avatar
IronMike
Black Belt - 2nd Dan
Posts: 2039
Joined: Thu May 12, 2016 6:13 am
Location: Boston
Languages: Russian, 3/3 (DLPT, 2021) 2+ (OPI, 2021)
Esperanto, C1 (KER skriba ekzameno, 2017)
Italian, 1L/2R (DLPT, 2019)
BCS, 3L/2+R/2S (DLPT in, oh God, 1999!)
Slovene, 2+L/3R (DLPT in, yes, 1999)
Language Log: viewtopic.php?f=15&t=5189
x 4986
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby IronMike » Thu Apr 15, 2021 12:57 am

tungemål wrote:I'm always wondering about this when writing e-mails in English. I have received "with best regards" from a sender in the UK. American e-mails seem to use "cheers". So what's the norm for formal e-mails? A formal e-mail is on par with a letter.

Le Baron wrote:I don't think I have ever written or received 'best regards' at the end of an email or a real letter. Who actually says 'best regards' and why? It was never standard English in the UK.

'Kind regards' (which I've actually received a few times) is another which seems to have a vague meaning to me. Do they mean 'best wishes'? Even that I reserve that for actual best wishes like a birthday or at a push to someone I know more informally. 'Regards' seemed to turn up when I was first using e-mail in the late 1990s and (in the UK at least) it collapsed into 'cheers' on the informality scale.

I get the feeling people don't really know how to sign off in an email because it's in a no man's land between a letter and 'chat'. :lol:

This whole series is interesting to me.

In the military, I had always signed off my emails with "Sincerely."

Then in 1999, I went to Officer Training School, after 13 years already in the service. 100+ days away from connected computers and official email.

Then I arrived at my first job. One of my additional duties was to organize graduations for officers in technical training. Part of my duty was to send out an email to all the senior officers (colonels) on base inviting them to the graduations (every three weeks or so).

A few months into this additional duty I was called into my captain's office and counseled on how to sign off an email. Apparently, during my OTS time, the entire air force changed how emails should be signed off. I was told I had to sign off emails with "v/r, Mike". Even worse, apparently I was only supposed to email officers one rank above me. (He was wrong on this, as my email to these colonels was "in the name of" my squadron commander, who was a lieutenant colonel.) Oh, and "v/r" means "very respectfully." A$$hole that I am, I asked my captain: Wouldn't it be more respectful to spell out very respectfully? He told me to get out of his office.

Oh, and guess what? The Navy has their own rules; if you, a sailor, are emailing someone your same rank or lower, you sign off with simply "r," Yep, the minute someone I'd been calling Joe for years got promoted above me, all of a sudden I would be expected to change my email to him to "v/r" Also, the minute a Navy junior to me would be promoted to my same rank, s/he would start signing off their emails with "r," whereas just a day earlier s/he would use "v/r,"

Stupid Navy.
3 x
You're not a C1 (or B1 or whatever) if you haven't tested.

User avatar
rdearman
Site Admin
Posts: 5602
Joined: Thu May 14, 2015 4:18 pm
Location: United Kingdom
Languages: English (N)
Language Log: viewtopic.php?f=15&t=1836
x 14811
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby rdearman » Thu Apr 15, 2021 5:22 pm

After getting out of the USAF I made a point of going to RAF Lakenheath with a retired buddy of mine and calling EVERY officer I saw over to me by saying "Hey Sarge, come here a second."

Oh how we laughed. What could they say to the civilian?

:D :lol:
3 x
Anyone who thinks assembly language programming is difficult, obviously hasn't used Rust.

The Autodidactic Podcast
The Lollygagging Podcast

User avatar
IronMike
Black Belt - 2nd Dan
Posts: 2039
Joined: Thu May 12, 2016 6:13 am
Location: Boston
Languages: Russian, 3/3 (DLPT, 2021) 2+ (OPI, 2021)
Esperanto, C1 (KER skriba ekzameno, 2017)
Italian, 1L/2R (DLPT, 2019)
BCS, 3L/2+R/2S (DLPT in, oh God, 1999!)
Slovene, 2+L/3R (DLPT in, yes, 1999)
Language Log: viewtopic.php?f=15&t=5189
x 4986
Contact:

Re: What does the (distant) future of language acquisition look like?

Postby IronMike » Thu Apr 15, 2021 9:45 pm

rdearman wrote:After getting out of the USAF I made a point of going to RAF Lakenheath with a retired buddy of mine and calling EVERY officer I saw over to me by saying "Hey Sarge, come here a second."

Oh how we laughed. What could they say to the civilian?

:D :lol:

Love this.

My favorite boss and probably the best intel analyst I ever knew (RIP, SSgt Jimmy Ryan), once gave me a memory I'll never forget. This was 33 years ago and I still remember it like it was yesterday.

We had just finished a mid shift (2200-0600) and we were all tired. We were on the shift bus waiting for one more person from our unit so we could get back to the dorms. SSgt Ryan was outside, leaning against the bus to keep from falling asleep on the bus, and had his uniform cap on the back of his head, like some farmer. An Army lieutenant came up the street going into work. Jimmy saw the Lt and gave him a lackluster salute, still leaning against the bus, cap still on the back of his head. At the same time, our last person arrived and got on the bus.

The lieutenant looked at Jimmy angrily and said, "I don't think that's how a salute should be rendered, Sergeant."

Jimmy, no sh!t, took off his cap and started getting on the bus, and said, "Sir, I just worked a mid. You're f^cking lucky you got that." Then the German contractor (bus driver) shut the bus's door and started driving away.

It could NOT have been more perfect.
1 x
You're not a C1 (or B1 or whatever) if you haven't tested.


Return to “General Language Discussion”

Who is online

Users browsing this forum: Kraut, Mr Dastardly and 2 guests