IPA mapping for Polish lexicon?

[brawer]

What is the correct mapping from allophones in the MaryTTS Polish lexicon to IPA?

For Unicode’s Unilex project, I’m trying to convert the Polish pronunciation dictionary from MaryTTS to the International Phonetic Alphabet. (Thanks again for collaborating with Unicode, you really have nice data!) In Polish, it seems MaryTTS uses a custom phonetic notation which seems to be mostly like X-SAMPA, but with custom modifications. For example, the X-SAMPA symbol y would be y in IPA, which isn’t a Polish vowel; probably MaryTTS uses y for the IPA vowel ɨ but I’m not fully sure.

Here’s a first attempt for a mapping table. If you could kindly tell me the correct mappings (especially for the rows with *****), I’ll gladly volunteer to add an ipa attribute to allophones.pl.xml similar to the pull requests I’ve sent you for Luxembourgish, German and French. Perhaps more importantly, we can then import your data into Unicode’s lexicon.

MaryTTS	Count	IPA
.	35427	.
o	8487	o
a	8429	a
e	7937	e
n	5625	n
r	5599	r
v	5382	v
t	5187	t
j	5067	j
aa	4850	*****
p	4687	p
s	4286	s
y	4046	*****
oo	3999	*****
m	3964	m
k	3783	k
ee	3712	*****
i	3347	i
d	3296	d
u	3226	u
l	3183	l
ni	2978	*****
w	2733	w
z	2585	z
g	2101	ɡ
Sz	2005	*****
ts	1988	t͡s
f	1883	f
b	1773	b
x	1580	x
ii	1527	*****
tSz	1501	*****
yy	1331	*****
uu	1322	*****
tsi	1273	*****
rZ	1165	*****
w_	1158	*****
dzi	1004	*****
si	976	*****
c	772	*****
dz	421	d͡z
N_	326	*****
zi	219	*****
gi	143	*****
j_	109	*****
drZ	39	*****

[psibre]

@brawer The lexicon was assembled from resources provided by @jolabachan. I have to admit I'm not sure about the relation to IPA -- maybe @jolabachan can help?

[jolabachan]

Hi! I wasn't the creator of the notation, but I will do my best. The double syllables are accented syllables. We don't have long and short syllables in Polish, but in this notation this is how they were marked. The "." means the end of a syllable. Here are some examples: kazał => k aa . z a w kazała => k a . z aa . w a krzyczeć => k Sz yy . tSz e tsi These are not always long vowels, but they are more prominent than the unaccented ones.

MaryTTS	Count	IPA
.	35427	.
o	8487	o
a	8429	a
e	7937	e
n	5625	n
r	5599	r
v	5382	v
t	5187	t
j	5067	j
aa	4850	***** a
p	4687	p
s	4286	s
y	4046	***** ɨ
oo	3999	***** o
m	3964	m
k	3783	k
ee	3712	***** e
i	3347	i
d	3296	d
u	3226	u
l	3183	l
ni	2978	***** ɲ
w	2733	w
z	2585	z
g	2101	ɡ
Sz	2005	***** ʃ
ts	1988	t͡s
f	1883	f
b	1773	b
x	1580	x
ii	1527	***** i
tSz	1501	***** t͡ʃ
yy	1331	***** ɨ
uu	1322	***** u
tsi	1273	***** t͡ɕ
rZ	1165	***** ʒ
w_	1158	***** w̃
dzi	1004	***** d͡ʑ
si	976	***** ɕ
c	772	***** c [as in word 'kitel', palatalised /k/]
dz	421	d͡z
N_	326	***** ŋ
zi	219	***** ʑ
gi	143	***** ɟ [as in word 'gips', palatalised /g/]
j_	109	***** j̃
drZ	39	***** d͡ʒ

Good luck! :-)

[psibre]

@jolabachan Thanks for the helpful clarifications!