
Can go wrong, what!
AI can translate between languages in real time as people speak, according to fresh research from Chinese search giant Baidu and Oregon State University in the US. Human interpreters need superhuman concentration to listen to speech and translate at the same time. There are, apparently, only a few thousand qualified …
Reminds me of the old joke about a crowd of people next to a building site in Berlin watching the foreman berating one of the workers. An English visitor is walking by and is bemused by all these people standing there intently, so he asks one of them what they are doing. The reply is "We are waiting for the verb!"
I'd like to see how a statistical analysis machine would handle that.
Might I suggest Jurafsky & Martin, Speech and Language Processing? It's a standard introductory textbook in the field.
After that, there are, of course, thousands of papers available in sources such as arXiv and the ACM DL on the topic of handling ambiguous natural-language parses with statistical techniques. Research in the area goes back at least as far as Shannon.
English is like that. How about prepositions such as "with", like when you just bought something from Ikea.
"I built a bookcase with a screwdriver."
"I built a bookcase with Susan."
Completely obvious to a native speaker but imagine some poor bastard trying to learn English.
(Of course being Ikea it wouldn't be a book case it would be some word with lots of ä and å, just to complicate matters.)
Russia has in interesting solution to this specific example, in that, with a screwdriver is rendered in the pure instrumental case, and with Susan, using a preposition and instrumental, to indicate you are not using Susan to build the bookcase, but that she is accompanying you in the action.
Time flies like an arrow. Fruit flies like a banana.
1. Time passes swiftly.
2. The time flies around an archery range like arrows.
3. To time the velocity of a fully laden fly, you time them like you time arrows.
Xerox had an English language understanding project and had no trouble with sentences like these. The software could come up with all of these parses but needed more context to disambiguate amongst them.
They had a natural language airline reservation system but had problems like the following:
"I would like a reservation between Los Angeles and San Francisco."
"I have a flight leaving at 12:00."
"Do you have one a little closer to 6:00?"
"I have a flight leaving at 12:01."
Actually, there was a flight at 6:00 but the client specified "a little closer" and not "a lot closer."
Oh, well.
The European Parliament allows members to use their own EU language, and a team of interpreters provides translations into the others. The members know that they have to pause for the interpreters. The result is that oratory is destroyed.
It is possible to simultaneously listen to the speaker with one ear and to one of the interpreters via a headphone; but that causes remarkable brain strain.
The European Parliament allows members to use their own EU language, and a team of interpreters provides translations into the others. The members know that they have to pause for the interpreters. The result is that oratory is destroyed.
Probably a good thing as the pauses let you think about what's said rather than just emotionally reacting to the oratory. Mussolini and Hitler might have had less of an impact if they'd had to wait for translators after every sentence.
Ken, they put up a sandstone statue of Mel Gibson at the Wallace Monument, with the word 'Freedom' chiselled in. Everyone loathed it and vandalised it. So they put up a plinth stating that the sculptor was disabled and we should tolerate the obscenity. Everyone kept on vandalising it. So they put a twelve foot high black steel cage around the statue - there are photos on the internet - which kind of make a mockery of the word 'freedom'.
I went under the cage and hacked Gibson's nose with a chisel. Then they finally removed it.
We were also planning on stealing the Wallace sword, again it wouldn't be the first time it was stolen, but we couldn't agree who would store it. It is still there, but have a look at it sometime. It is taller than Gibson.
I fail to see this is any advance in Kevin Knight's paper from 1999
http://mt-archive.info/JHU-1999-AlOnaizan.pdf
With enough training data you get a better idea of what word follows another or group of others.
The fun comes with languages with relatively free word order i.e. where subject and objects carry case markers rather than by their position e.g. ones with lots of morphology.
Taking an example from the article, I would guess the AI would do fairly well with sentence fragments as it uses predictive analysis for much of its output.
"Translating between Japanese and German to English and Chinese, therefore, more difficult"
Would it fill in the implied "is"?
On page 6 of the paper I put a link to it talks about this.
If you have a series of matched translations and the Russian ones generally have fewer words than the English ones which contains "is" the model "realises" in the Russian copular is implied and so deasl with it.
The same goes for definite and indefinite articles.
Clever eh!
I remember hearing on the radio an interview with an English/Russian interpreter. During a meeting between an American president and his Russian counterpart, he used the expression “not buying pig in a poke”. This was translated into the Russian equivalent “not buying cat in a bag”. The Russian president replied by talking about cats.
I can’t remember how the interpreter solved the problem but it would be interesting to get AI to solve it, assuming it had an understanding of idiomatic phrases.
The comment by FlossyThePig put me in mind of an ancient joke...
A university had just developed a English/Chinese translator. During the dedication ceremony, one of the dignitaries demonstrated the computer by typing in the phrase "Out of sight, out of mind". The machine whirred, clicked, and produced a sheet of paper with Chinese characters on it. Unfortunately, the developers of the machine were absent, and no one at the dedication could read Chinese. So the dignitary told the computer to re-translate what it had output back to English. Again the machine whirred and clicked, and this time produced a page with "Invisible idiot" written upon it.
Yes, it's a very old joke.
>Wait a minute.... lets test it first and see. Mines a £50 bet that it cant differentiate.
I would happily bet a lot more than that !
As a regular visitor to Japan, I gave up a long time ago on "blackbox" translators (the sort trash Google translate puts out is laughable .... its cringeworthy to the English-speaking eye, let alone what it comes up with the other way around according to my Japanese friends).
So I have resorted to keeping a number of plain-old Japanese dictionary apps on my phone, along with embarking on that long journey of learning the Kanji. I know I will never speak the language to any degree of proficiency, but at least I can recognise words and some typical common phrase constructions, and I can take a reasonably well educated guess at what a restaurant menu is telling me ... so at least I won't starve and I get to experience a wider variety of delicious Japanese food.
"Technically, for German, it's V2 and SOV."
So, you mean it's 'second idea', like I said?
Well, technically, German has a "+Scramble" feature on its universal syntax. My personal favourite feature is +Scramble. Which means your trees are built up in the correct way, and then the words can be moved around afterwards.
The China Syndrome used to be about our inventions surfacing on the other side after it all had gone wrong on our side.
The modern version would be about the west inventing the basics and then finding AI being used by the super-state to control citizens in ways unimaginable only a few years ago. In fact, with their taking of Tibet, aspirations in the South Asian waters and internment of muslims we may expect them on our shores one day. :-(
Microsoft has pledged to clamp down on access to AI tools designed to predict emotions, gender, and age from images, and will restrict the usage of its facial recognition and generative audio models in Azure.
The Windows giant made the promise on Tuesday while also sharing its so-called Responsible AI Standard, a document [PDF] in which the US corporation vowed to minimize any harm inflicted by its machine-learning software. This pledge included assurances that the biz will assess the impact of its technologies, document models' data and capabilities, and enforce stricter use guidelines.
This is needed because – and let's just check the notes here – there are apparently not enough laws yet regulating machine-learning technology use. Thus, in the absence of this legislation, Microsoft will just have to force itself to do the right thing.
Comment More than 250 mass shootings have occurred in the US so far this year, and AI advocates think they have the solution. Not gun control, but better tech, unsurprisingly.
Machine-learning biz Kogniz announced on Tuesday it was adding a ready-to-deploy gun detection model to its computer-vision platform. The system, we're told, can detect guns seen by security cameras and send notifications to those at risk, notifying police, locking down buildings, and performing other security tasks.
In addition to spotting firearms, Kogniz uses its other computer-vision modules to notice unusual behavior, such as children sprinting down hallways or someone climbing in through a window, which could indicate an active shooter.
In brief US hardware startup Cerebras claims to have trained the largest AI model on a single device powered by the world's largest Wafer Scale Engine 2 chip the size of a plate.
"Using the Cerebras Software Platform (CSoft), our customers can easily train state-of-the-art GPT language models (such as GPT-3 and GPT-J) with up to 20 billion parameters on a single CS-2 system," the company claimed this week. "Running on a single CS-2, these models take minutes to set up and users can quickly move between models with just a few keystrokes."
The CS-2 packs a whopping 850,000 cores, and has 40GB of on-chip memory capable of reaching 20 PB/sec memory bandwidth. The specs on other types of AI accelerators and GPUs pale in comparison, meaning machine learning engineers have to train huge AI models with billions of parameters across more servers.
In Brief No, AI chatbots are not sentient.
Just as soon as the story on a Google engineer, who blew the whistle on what he claimed was a sentient language model, went viral, multiple publications stepped in to say he's wrong.
The debate on whether the company's LaMDA chatbot is conscious or has a soul or not isn't a very good one, just because it's too easy to shut down the side that believes it does. Like most large language models, LaMDA has billions of parameters and was trained on text scraped from the internet. The model learns the relationships between words, and which ones are more likely to appear next to each other.
Opinion The Turing test is about us, not the bots, and it has failed.
Fans of the slow burn mainstream media U-turn had a treat last week.
On Saturday, the news broke that Blake Lemoine, a Google engineer charged with monitoring a chatbot called LaMDA for nastiness, had been put on paid leave for revealing confidential information.
In the latest episode of Black Mirror, a vast megacorp sells AI software that learns to mimic the voice of a deceased woman whose husband sits weeping over a smart speaker, listening to her dulcet tones.
Only joking – it's Amazon, and this is real life. The experimental feature of the company's virtual assistant, Alexa, was announced at an Amazon conference in Las Vegas on Wednesday.
Rohit Prasad, head scientist for Alexa AI, described the tech as a means to build trust between human and machine, enabling Alexa to "make the memories last" when "so many of us have lost someone we love" during the pandemic.
Google has placed one of its software engineers on paid administrative leave for violating the company's confidentiality policies.
Since 2021, Blake Lemoine, 41, had been tasked with talking to LaMDA, or Language Model for Dialogue Applications, as part of his job on Google's Responsible AI team, looking for whether the bot used discriminatory or hate speech.
LaMDA is "built by fine-tuning a family of Transformer-based neural language models specialized for dialog, with up to 137 billion model parameters, and teaching the models to leverage external knowledge sources," according to Google.
Analysis After re-establishing itself in the datacenter over the past few years, AMD is now hoping to become a big player in the AI compute space with an expanded portfolio of chips that cover everything from the edge to the cloud.
It's quite an ambitious goal, given Nvidia's dominance in the space with its GPUs and the CUDA programming model, plus the increasing competition from Intel and several other companies.
But as executives laid out during AMD's Financial Analyst Day 2022 event last week, the resurgent chip designer believes it has the right silicon and software coming into place to pursue the wider AI space.
GPUs are a powerful tool for machine-learning workloads, though they’re not necessarily the right tool for every AI job, according to Michael Bronstein, Twitter’s head of graph learning research.
His team recently showed Graphcore’s AI hardware offered an “order of magnitude speedup when comparing a single IPU processor to an Nvidia A100 GPU,” in temporal graph network (TGN) models.
“The choice of hardware for implementing Graph ML models is a crucial, yet often overlooked problem,” reads a joint article penned by Bronstein with Emanuele Rossi, an ML researcher at Twitter, and Daniel Justus, a researcher at Graphcore.
As compelling as the leading large-scale language models may be, the fact remains that only the largest companies have the resources to actually deploy and train them at meaningful scale.
For enterprises eager to leverage AI to a competitive advantage, a cheaper, pared-down alternative may be a better fit, especially if it can be tuned to particular industries or domains.
That’s where an emerging set of AI startups hoping to carve out a niche: by building sparse, tailored models that, maybe not as powerful as GPT-3, are good enough for enterprise use cases and run on hardware that ditches expensive high-bandwidth memory (HBM) for commodity DDR.
Biting the hand that feeds IT © 1998–2022