Tone changes gives error in tone practice

Aslaksen

秀才
I'm using the tone practice, and I find it very useful. However, would it be possible for you to not count it as an error when I answer 23 when they read 33? I go by what I hear. Thanks!
 

Shun

状元
Hi Pierre,
Hi Aslaksen,

I’m not sure if it's always a good idea for Pleco to ask users to take tone changes into account when entering tones in flashcard tests. A user should know both the lexical tone of each syllable and the tone sandhi rules, which are quite simple to internalize and shouldn't really be practiced by entering them into a smartphone, but instead by pronouncing them. If Pleco asks for the tone as it's spoken, there's a definite risk that a user won't learn the actual, lexical tone properly. Also in cases like the following, it would quickly get a little impenetrable, especially for new users, if Pleco asked for the changed tone:
  • Tone changes of 一 yī, where the first tone can sometimes turn into a second, and sometimes into a fourth tone
  • Series of tone changes, such as 我很好!, where both the 我 and the 很 have their tone changed from third to second.
More importantly, it may not be all that easy to program. What if you have four third tones lexically, then Pleco would have to change three syllables' tones, which usually isn't right. It can quickly get a little complicated for a user to understand clearly whether a syllable's tone has changed or whether that's its original.

In addition, the real-life tone changes are often more complicated than the relatively simple tone sandhi rules we find in most textbooks. So it may be teaching you something simple which is actually wrong (or strangely super-correct) in many cases. A new grammar book by Jeroen Wiedenhof, which breaks with principles posited by many grammars, details this, as well:


Of course, it's still a matter of personal preference, and Pleco could also add an option for it. But when we don't want to do that, I feel that asking for lexical tones is the safer route.

In the case of longer expressions in dictionaries, it commonly happens that tone changes are incorporated in the expression. In this case Pleco would already ask for those changed tones, and there it makes more sense than in the case of disparate words put together.

Cheers, and good studying,

Shun
 

Aslaksen

秀才
Hi Shun,

Can I just double check: You do realize that we are talking about the tone tests, and not flashcard tests? We don't see the characters before we have answered. Maybe I misunderstand you, but when I read your reply, it sounds to me like you are talking about flashcard tests where I see the characters and then enter the tones. I agree with you that flashcard tests should not use tone changes. But if Pierre simply hears ying2 xiang3, do you then expect him to know the word, and answer 33? What if he doesn't know the word? How could he possible get the question right? One of the things I like about the Pleco tone test, is that it doesn't use simple words. A lot of other tone tests use words that I immediately recognize, and that helps me know the tones. I prefer tests that use names or more complicated words, so that I have to rely exclusively on my listening.
 

Shun

状元
Hi Aslaksen,

thanks for your reply. Yes, I'm aware that it's the test where you can tap the tone numbers for a word you see or hear. You have a point if you just want to train your tone recognition with words you perhaps haven't ever seen before (I've never done it like that.). That's indeed a very different use case. I had always assumed the Tone practice feature in flashcards to be there for you to train your memory of a word's tones.

In this case the question of how to implement the tone practice feature will depend on how many users do it your way, and what percentage of users do it my way. Perhaps @mikelove knows more here, so I'll leave this question to him.

Cheers, Shun
 
Last edited:

Aslaksen

秀才
Hi Shun,

I'm sorry, but I still feel that you don't understand what I'm talking about. If you go to Profile and select Tones (instead of Simple or Spaced Repetition) and under Test type select Tone Practice, you get a test where you have no control of which words you get. So unless you're a native speaker, many/most of the words will be unknown. You talk about "your way" and "my way", but what we're talking about is really "your test" (flashcard where you want to test if you remember the tones) or "my test" (testing my ability to recognize tones from listening). I assume that you choose Simple or Spaced Repetition in Profile. Since that will be a fundamentally different type of test, there is no reason why the rules should be the same in the two cases. The choice between your way or my way has already been made when we choose profile.
 

Shun

状元
Hi Aslaksen,

the thing is, if you choose a different profile, that isn't a setting in itself, but instead in your case, it switches around a lot of other settings that you will then perceive as the app being in a different mode. These settings also include which flashcards are going to be chosen for the test. Here, it's probably just all flashcards when you choose your "Tones" profile. This is why you may believe the set of words is completely random and unknown. Also, in your Tones profile, you have "Show" set to "Audio". That's the whole magic behind it. I'm 99.9% sure.

So I agree with you that with your current settings geared towards tone recognition, the tones Pleco asks for might preferably include tone changes, different from those settings which are geared towards memorization of tones from reading the Hanzi.

Hope this clears it up, Shun
 

Aslaksen

秀才
Yes, I agree. Tone changes or not should be determined by the choice of profile. And the reply from Mike Love to my original post indicates that that is the intention. I'm just trying to point out that it doesn't seem to have been implemented yet.
 
More importantly, it may not be all that easy to program. What if you have four third tones lexically, then Pleco would have to change three syllables' tones, which usually isn't right. It can quickly get a little complicated for a user to understand clearly whether a syllable's tone has changed or whether that's its original.
I don’t fully agree: Pleco already pronounces with certain tones, I just want to practice what I hear (wether it’s correct or not). If I just want to train the individual character tone, then there’s no point in listening I guess (unless you just want to test whether the first character is a real 2 or a changed 3 due to 3-3)
 

Shun

状元
I don’t fully agree: Pleco already pronounces with certain tones, I just want to practice what I hear (wether it’s correct or not). If I just want to train the individual character tone, then there’s no point in listening I guess (unless you just want to test whether the first character is a real 2 or a changed 3 due to 3-3)

Yep, so we are in line, since the test you're describing is also tone recognition, unlike the different case of tone memorization. Let's wait and see which type Mike thinks is more common. It would definitely be nice to somehow have both.
 

mikelove

皇帝
Staff member
Yeah, we looked at this and ended up deciding to wait until we roll out more robust tone sandhi support in general (adding the option to display / search for it as well as using it for audio like we do now) - didn't want to roll out a half-backed hack-y version of it now and then potentially have the test behavior change again when we do that, since any change to which answers the system deems correct is necessarily going to be disruptive for a lot of people who suddenly find that formerly-incorrect answers they trained around are now correct or formerly-correct answers are now incorrect.
 

Shun

状元
Thanks! Yeah, authentic tone sandhi support is absolutely non-trivial—with good example data, it might even benefit from machine learning. But I'm guessing that since iPhones and Android devices aren't equally far along regarding built-in hardware machine learning support, offering this cross-platform in Pleco would only be possible in a couple of years from now.
 

mikelove

皇帝
Staff member
Nothing that complicated, actually, it's just a matter of having a dedicated sandhi field in databases. (as with Cantonese e.g.) We're also adding a Taiwan Mandarin one at the same time, actually (and with its own sandhi version), to avoid the awkward way that's currently implemented with Cross-Straits.
 

Shun

状元
Thanks, I'll be curious to see how it's possible by means of a fixed data field since words can be freely combined, but of course I fully believe it when you say it.
 

mikelove

皇帝
Staff member
It'd be a field on flashcards too, so if you make a custom flashcard out of multiple words it'll take a stab at automatically generating the sandhi for you but you can then customize it to fix any mistakes.
 

Shun

状元
Thanks, that sounds nice. Since users often don't know the right sandhi, in this area I thought machine learning might come in handy, though I understand this may also take more research on the linguistic side to get all the authentic example data gathered together. Perhaps it's also completely logically describable, in which case ML would indeed be unnecessary.
 

Aslaksen

秀才
Yeah, we looked at this and ended up deciding to wait until we roll out more robust tone sandhi support in general (adding the option to display / search for it as well as using it for audio like we do now) - didn't want to roll out a half-backed hack-y version of it now and then potentially have the test behavior change again when we do that, since any change to which answers the system deems correct is necessarily going to be disruptive for a lot of people who suddenly find that formerly-incorrect answers they trained around are now correct or formerly-correct answers are now incorrect.
I'm sorry, but I don't understand. Is your concern that some people who are good in Chinese may recognize that what I hear as 23 is actually a 33 and will answer 33? First of all, who used the tone test profile? I assume that is mostly people who are not good at Chinese, and who are therefore not likely to be sure that what they hear as 23 is actually 33. And secondly, can't you just state up front that this is a "what you hear is what you get" test? I totally get that it would be very hard to do tone changes in a flash card test, but that is not at all what Pierre and I are talking about. All we are asking about is to not be marked as wrong when we actually detect the right tone.
 
Top