Hacker News Reader

Comment by walterbell | Hacker News Reader

Alexa currently requires memorization of command sequences. Can take a while for older users to learn.

E.g. there's a skill (AskMyBuddy) that sends a preprogrammed text to a list of cellphones, typically a request for help. But the user needs to be trained to remember the exact command ("Alexa Ask My Buddy to send help"), which might be forgotten if someone is in a stressful situation.

Anecdote: tried to create a todo list with one item, "buy milk", but Alexa would not accept this item unless you setup the ability to purchase items from Amazon.

Weather does not work outside the US (needs a US address).

View on HN

However, its NLP is quite good. Having my (Vietnamese-born) parents try the Echo Dot, I was struck by how awful their English actually is, at least in comparison to how my brain has come accustomed to their speech patterns. For one thing, they enunciate "Alexa" much differently than I do. And then where I would say, "What's the weather today", they say, "What weather today is?" or "What day this week will be snow?" To my surprise, Alexa actually understood their convoluted phrasing, at least for the built-in skills that Alexa has had well-honed out of the box.

Thanks for this comment, I hadn't really thought about how NLP may fail ESL speakers.

Do you think your parents would be more comfortable talking to it in Vietnamese if it was possible? I'm mostly wondering if handling good Vietnamese is easier than handling English with poor grammar.

Definitely. Their English even after 30+ years in America is at the level where our English-English conversations are probably at the elementary grade level (except with a few more proper nouns). I used to think that they were at least better in understanding my natural English conversation, but I've realized that I reflexively shift to using much simpler sentences when speaking to them. But when they're with Vietnamese friends or family, they have conversations (in Vietnamese) just as normal adults typically do in their native language.

Conversely, I've since realized that the Vietnamese that I think I understand is probably at the toddler level. For a very long time, I just assumed the Vietnamese language lacked features such as pronouns and articles. But then I realized that when my parents tell me to go wash the dishes, my brain just fixates on "wash dishes" and ignores all the other connective words. So I know a lot of verbs and nouns but very few words that are part of everyday conversational speech in Vietnamese. I imagine that's what Alexa feels like :)

Your comment made me curious how well the Google Translate app can deal with foreign languages. I was stunned to see that it could understand my attempts at Vietnamese. So other than a proportionately smaller dataset to learn from (Vietnamese usage vs English usage of Google or Amazon), seems like Alexa and Google Home could competently deal with foreign languages.

Handling poor English grammar is a necessity anyway if you want to hit mass market share, since not everyone speaks as if they read the Oxford English Dictionary every night even when they are native English speakers.

"What weather today is" Why is this considered an amazing NLP? I type any order of related words in Google (or any other search engine) and the intend of my malformed sentence is sufficiently understood to bring back relevant results. Even if I formed my sentence grammatically, the result is no better.

That's because most search engines use a form of NLP in their processing of user input for queries.

No. the most trivial search strategy just tokenizes the input and looks up those token in the index. A poorly formed sentence, is not a detriment to finding out the intend, as long as key words are there. It doesn't require complex NLP to achieve this end result.

Ironically, "What day this week will be snow?" doesn't give me great results on Google, but Bing actually shows a weekly weather graph.

"snow this week" will return the weather forecast for my location on google

Yeah, but that's grammatically valid. It's not a question, but it's a valid sentence.

I don't think it's a valid sentence, unless you interpret it as commanding someone/something to snow sometime this week.

So will "this snow week when"

Then why is it that you get just as good, if not better, search results when you type words in decreasing order of information content and omit common words, some of which are essential for parsing natural languages? E.g.

   Beatles walrus -> the song "I am the walrus" by the Beatles.
   Tintern vacuum -> the song "Vacuum cleaner" by Tintern Abbey.

If you say the intent activation word (weather / snow / light etc..) the intent will trigger. Phrasing can help for certain queries though.

bingo!

Interesting. The phrases I typically use for Google Search are also not considered proper English grammar, but are highly effective in getting the results that I want. I wonder if that's the case as well here.

Most commands are pretty obvious and easy, but some of them aren't. For instance, the phone locator. The command is something like, "Alexa, tell trackr to locate my phone." But my wife and I can never remember that when we need to locate a phone, and we have to figure out it out all over again.

I assume that people have to call for emergency services less often than we misplace our phones, so I'm sure it's even harder to remember the exact verbiage for that.

Alexa added "Ham Bread" to my shopping list yesterday.

Not super on-topic, but I laughed when I saw it.

Weather works quite well for me here in germany.

Thanks. Looks like some countries work, others don't.

Weather works fine in the UK.

Reply on news.ycombinator.com