예제 #1
0
def from_bibkey(glottocode, bibkey):
	"""Look up source info from the Glottolog bibkey of the source.
	You'll probably want to change GLOTTOLOG_PATH here."""
	GLOTTOLOG_PATH = path.expanduser('~/Documents/glottolog-4.0')

	from pyglottolog import Glottolog
	g = Glottolog(GLOTTOLOG_PATH)

	languoid = g.languoid(glottocode)	
	ref = [x for x in languoid.sources if x.key == bibkey]
	if not ref:
		raise Exception('Reference not found')

	source = ref[0].get_source(g)
	return source.fields
예제 #2
0
def main(args):
    assert args.glottolog, 'The --glottolog option is required!'
    license = licenses.find(args.cldf.properties['dc:license'])
    assert license and license.id.startswith('CC-')

    data = Data()
    ds = data.add(
        common.Dataset,
        mixezoqueanvoices.__name__,
        id=mixezoqueanvoices.__name__,
        name="Mixe-Zoquean Voices",
        domain='mixezoqueanvoices.clld.org',
        publisher_name="Max Planck Institute for Evolutionary Anthropology",
        publisher_place="Leipzig",
        publisher_url="http://www.eva.mpg.de",
        license=license.url,
        jsondata={
            'license_icon':
            '{}.png'.format('-'.join(
                [p.lower() for p in license.id.split('-')[:-1]])),
            'license_name':
            license.name
        },
    )

    contrib = data.add(
        common.Contribution,
        None,
        id='cldf',
        name=args.cldf.properties.get('dc:title'),
        description=args.cldf.properties.get('dc:bibliographicCitation'),
    )
    data.add(common.Contributor, 'kondic', id='kondic', name='Ana Kondic')
    data.add(common.Contributor, 'gray', id='gray', name='Russell Gray')
    DBSession.add(
        common.ContributionContributor(
            contribution=contrib,
            contributor=data['Contributor']['kondic'],
        ))
    for i, ed in enumerate(['kondic', 'gray']):
        data.add(common.Editor,
                 ed,
                 dataset=ds,
                 contributor=data['Contributor'][ed],
                 ord=i)

    ancestors = collections.defaultdict(list)
    gl = Glottolog(args.glottolog)
    lnames = {}
    for lang in args.cldf.iter_rows('LanguageTable', 'id', 'glottocode',
                                    'name', 'latitude', 'longitude'):
        lnames[lang['id']] = lang['name']
        glang = None
        if lang['glottocode']:
            glang = gl.languoid(lang['glottocode'])
            lineage = [i[0] for i in glang.lineage]
            if 'Mixe-Zoque' in lineage:
                ancestors[lang['id']].append('Protomixezoque')
            if 'Mixe' in lineage:
                ancestors[lang['id']].append('Protomixe')
            if 'Oaxaca Mixe' in lineage:
                ancestors[lang['id']].append('Protooaxacamixe')
        if not glang:
            assert lang['name'] == 'Nizaviguiti'
        data.add(
            models.Variety,
            lang['id'],
            id=lang['id'],
            name=lang['name'],
            latitude=lang['latitude'],
            longitude=lang['longitude'],
            glottocode=lang['glottocode'],
            description=lang['LongName'],
            subgroup=glang.lineage[1][0]
            if glang and len(glang.lineage) > 1 else None,
        )
    colors = dict(
        zip(
            set(l.subgroup for l in data['Variety'].values()),
            qualitative_colors(
                len(set(l.subgroup for l in data['Variety'].values())))))
    for l in data['Variety'].values():
        l.jsondata = dict(color=colors[l.subgroup].replace('#', ''))

    for rec in bibtex.Database.from_file(args.cldf.bibpath, lowercase=True):
        data.add(common.Source, rec.id, _obj=bibtex2source(rec))

    refs = collections.defaultdict(list)

    # Store proto-forms for later lookup:
    proto_forms = collections.defaultdict(
        lambda: collections.defaultdict(list))
    for form in args.cldf.iter_rows('FormTable', 'id', 'form',
                                    'languageReference', 'parameterReference'):
        if form['languageReference'].startswith('Proto'):
            proto_forms[form['languageReference']][
                form['parameterReference']].append(form['form'])

    for param in args.cldf.iter_rows('ParameterTable', 'id',
                                     'concepticonReference', 'name'):
        proto = collections.OrderedDict()
        for lid, forms in proto_forms.items():
            f = forms.get(param['id'])
            if f:
                proto[lnames[lid]] = f
        data.add(
            models.Concept,
            param['id'],
            id=param['id'],
            name='{} [{}]'.format(param['name'], param['id'].split('_')[0]),
            concepticon_id=param['concepticonReference'],
            concepticon_gloss=param['Concepticon_Gloss'],
            description=param['Spanish_Gloss'],
            jsondata=dict(reconstructions=proto),
        )

    f2a = form2audio(args.cldf)
    for form in args.cldf.iter_rows('FormTable', 'id', 'form',
                                    'languageReference', 'parameterReference',
                                    'source'):
        assert not (form['form'] == '►' and not f2a.get(form['id']))
        vsid = (form['languageReference'], form['parameterReference'])
        vs = data['ValueSet'].get(vsid)
        if not vs:
            vs = data.add(
                common.ValueSet,
                vsid,
                id='-'.join(vsid),
                language=data['Variety'][form['languageReference']],
                parameter=data['Concept'][form['parameterReference']],
                contribution=contrib,
            )
        for ref in form.get('source', []):
            sid, pages = Sources.parse(ref)
            refs[(vsid, sid)].append(pages)
        proto = collections.OrderedDict()
        for lid in ancestors.get(form['languageReference'], []):
            f = proto_forms[lid].get(form['parameterReference'])
            if f:
                proto[lnames[lid]] = f
        data.add(
            Counterpart,
            form['id'],
            id=form['id'],
            name=form['form'],
            valueset=vs,
            audio=f2a.get(form['id']),
            jsondata=dict(reconstructions=proto),
        )

    for (vsid, sid), pages in refs.items():
        DBSession.add(
            common.ValueSetReference(valueset=data['ValueSet'][vsid],
                                     source=data['Source'][sid],
                                     description='; '.join(nfilter(pages))))