Skip to content
This repository has been archived by the owner on Jan 2, 2021. It is now read-only.
/ conjugar Public archive

a helper for learning-spanish verbs, for exzample,the "conjugado" of spanish verbs

License

Notifications You must be signed in to change notification settings

ihgazni2/conjugar

Repository files navigation

  • pip3 install conjugar
  • some words include y need special handling,maybe not correct

  • get silabas

  • essila <word>

    CONJUGAR# essila desayunar
    -----------------------------
    de-sa-yu-nar
    -----------------------------
    {
     'silabas':
                [
                 'de',
                 'sa',
                 'yu',
                 'nar'
                ],
     'accent_pos': 3,
     'accent': 'nar',
     'stress_char_pos': 1,
     'stress_char': 'a'
    }
    -----------------------------
    ['de', 'sa', 'yu', 'nar']
    CONJUGAR#
    
  • esverb <string-pattern>

    CONJUGAR# esverb yu
    ayunar
    desayunar
    subyugar
    yugular
    yuxtaponer
    ayudar
    
  • in progressing(some special words out of rule)....

  • esconju <word> <ip ipi ipps ifi cs sp sfi spi ia>

    CONJUGAR# esconju registrar ip sp
    
            'ip':'indicativo presente',
            'ipps':'indictivo pretérito perfecto simple',
            'ipa':'indicativo pretérito anterior',
            'ipi':'indicativo pretérito imperfecto',
            'ifi':'indicativo futuro imperfecto',
            'sp':'subjuntivo presente',
            'spi':'subjuntivo pretérito imperfecto',
            'sfi':'subjuntivo futuro imperfecto',
            'cs':'condicional simple',
            'ia':'imperativo afirmativo',
    
    ip ipi ipps ifi cs sp sfi spi ia
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |         person|                   ip|         sp|prsn_abbr|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |             yo|             registro|   registre|       s1|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |         tú/vos|registras / registrás|  registres|       s2|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |          usted|             registra|   registre|      s23|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |        él,ella|             registra|   registre|       s3|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |nostros,nostras|          registramos|registremos|      pl1|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |vostros,vostras|           registráis| registréis|      pl2|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |        ustedes|            registran|  registren|     pl23|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    |    ellos,ellas|            registran|  registren|      pl3|
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    
from conjugar import acentuacion as acen
from xdict.jprint import pobj,pdir
  • show_stress

conjugar/Images/show_stress.0.png

  • get_stress

conjugar/Images/get_stress.0.png

  • get_silabas

    s = 'paraguayo'
    acen.get_silabas(s)
    >>>['pa', 'ra', 'guay', 'o']
    
  • get_accent_pos

    s = "paraguayo"
    acen.get_accent_pos(s)
    >>>(2,"guay")
    
  • get_spans

    spans = acen.get_spans(silabas)
    spans
    >>>>[(0, 2), (2, 4), (4, 8), (8, 9)]
    
  • word2clstarr

    acen.word2clstarr(s)
    >>>['p', 'a', 'r', 'a', 'g', 'uay', 'o']
    
  • AC_A

  • AC_E

  • AC_I

  • AC_O

  • AC_U

    >>> acen.AC_A
    

    'á' >>> acen.AC_E 'é' >>> acen.AC_I 'í' >>> acen.AC_O 'ó' >>> acen.AC_U 'ú' >>>

  • CE_C

    acen.CE_C
    >>>>'ç'
    
  • DI_U

    acen.DI_U
    >>>'ü'
    
  • ENE

    acen.ENE
    >>>'ñ'
    
  • UD_EXCM

    acen.UD_EXCM
    >>>'¡'
    
  • UD_QM

    acen.UD_QM
    >>>'¿'
    
  • ALPHABETA

    acen.ALPHABETA
    >>>'abcdefghijklmnopqrstuvwxyzáéíóúñü!¡?¿ç'
    
  • ACUTE_CHARS

    acen.ACUTE_CHARS
    >>>
    'áéíóú'
    
  • ACUTE_CHARS_MIRROR_DICT

    pobj(acen.ACUTE_CHARS_MIRROR_DICT)
    >>>
    {
        'a': 'á',
        'e': 'é',
        'i': 'í',
        'o': 'ó',
        'u': 'ú',
        'á': 'a',
        'é': 'e',
        'í': 'i',
        'ó': 'o',
        'ú': 'u'
    }
    
  • VOWEL

    acen.VOWEL
    >>>['a', 'e', 'i', 'o', 'u', 'á', 'é', 'í', 'ó', 'ú']
    
  • VOWEL_CHARS

    acen.VOWEL_CHARS
    >>>'aeiou'
    
  • STRONG_VOWEL_CHARS

    acen.STRONG_VOWEL_CHARS
    >>>'aeo'
    
  • WEAK_VOWEL_CHARS

    acen.WEAK_VOWEL_CHARS
    

' >>>'iu'

  • NORMAL_STRONG_VOWEL

    acen.NORMAL_STRONG_VOWEL
    >>>['a', 'e', 'o']
    
  • NORMAL_VOWEL

    acen.NORMAL_VOWEL
    >>>['a', 'e', 'i', 'o', 'u']
    
  • NORMAL_WEAK_VOWEL

    acen.NORMAL_WEAK_VOWEL
    >>>['i', 'u']
    
  • ACUTE_STRONG_VOWEL

    acen.ACUTE_STRONG_VOWEL
    >>>['á', 'é', 'ó']
    
  • ACUTE_VOWEL

    acen.ACUTE_VOWEL
    >>>['á', 'é', 'í', 'ó', 'ú']
    
  • ACUTE_WEAK_VOWEL

    acen.ACUTE_WEAK_VOWEL
    >>>['í', 'ú']
    
  • BI_CONSONANT

    acen.BI_CONSONANT
    >>>['ch', 'll', 'rr', 'pl', 'bl', 'tl', 'dl', 'cl', 'gl', 'pr', 'br', 'tr', 'dr', 'cr', 'gr']
    
  • CL_CONSONANT

    acen.CL_CONSONANT
    >>>['pl', 'bl', 'tl', 'dl', 'cl', 'gl', 'pr', 'br', 'tr', 'dr', 'cr', 'gr']
    
  • DI_CONSONANT

    acen.DI_CONSONANT
    >>>['ch', 'll', 'rr']
    
  • CONSONANT

    acen.CONSONANT
    >>>['b', 'c', 'd', 'f', 'g', 'h', 'j', 'k', 'l', 'm', 'n', 'p', 'q', 'r', 's', 't', 'v', 'w', 'x', 'y', 'z', 'ñ']
    
  • DIPTRONGO

    acen.DIPTRONGO
    >>>['ai', 'au', 'ei', 'eu', 'oi', 'ou', 'ia', 'ie', 'io', 'ua', 'ue', 'uo', 'iu', 'ui', 'ái', 'áu', 'éi', 'éu', 'ói', 'óu', 'iá', 'uá', 'ié', 'ué', 'ió', 'uó', 'üe', 'üi']
    
  • TRIPTONGO

    acen.TRIPTONGO
    >>> ['iai', 'iau', 'iei', 'ieu', 'ioi', 'iou', 'iái', 'iáu', 'iéi', 'iéu', 'iói', 'ióu', 'uai', 'uau', 'uei', 'ueu', 'uoi', 'uou', 'uái', 'uáu', 'uéi', 'uéu', 'uói', 'uóu', 'üei', 'üeu']
    
  • Y_DIPTRONGO

    acen.Y_DIPTRONGO
    >>>['ay', 'ey', 'oy', 'uy', 'áy', 'éy', 'óy', 'üy']
    
  • Y_TRIPTONGO

    acen.Y_TRIPTONGO
    >>>['iay', 'iey', 'ioy', 'iáy', 'iéy', 'ióy', 'uay', 'uey', 'uoy', 'uáy', 'uéy', 'uóy', 'üey']
    
  • Y_LAST

    acen.Y_LAST
    >>>['@']
    
  • is_vowclst

    >>> acen.is_vowclst("a")
    True
    >>> acen.is_vowclst("ai")
    True
    >>> acen.is_vowclst("iai")
    True
    >>> acen.is_vowclst("aa")
    False
    >>> acen.is_vowclst("bi")
    False
    >>>
    
  • is_conclst

  • is_biconstant

  • is_diptrongo

  • is_triptongo

  • is_acute_char

    acen.is_acute_char('a')
    >>>False
    acen.is_acute_char('á')
    >>>True
    
  • to_acute_char

    acen.to_acute_char('a')
    >>>'á'
    
  • to_non_acute_char

    acen.to_non_acute_char('a')
    >>>'a'
    
  • acute

    acen.acute('abcde')
    >>>'ábcdé'
    
  • deacute

    acen.deacute('ábcdé')
    >>>'ábcdé'
    
  • de_y

    s = "paraguayo"
    de_y_rslt = acen.de_y(s)
    de_y_rslt
    >>>['paragu', 'ayo']
    
  • de_trip

  • de_dip

  • de_bi

  • de_c

  • de_v

    internal using,  de_y->de_trip->de_dip->de_bi->de_c->de_v
    
  • de_engine

    internal used by  de_y,de_trip,de_dip,de_bi,de_c,de_v
    
  • arr_recvr_lasty

  • arr_repl_lasty

  • word_recvr_lasty

  • word_repl_lasty

  • get_stress_charpos_of_silaba

    >>> acen.get_stress_char_pos_of_silaba("guay")
    2
    >>> acen.get_stress_char_pos_of_silaba("no")
    1
    >>> acen.get_stress_char_pos_of_silaba("o")
    0
    >>>
    
  • get_charloc

  • get_spanloc

  • get_spans

  • acute_vowel_in

  • Help

  • In Progressing.....

About

a helper for learning-spanish verbs, for exzample,the "conjugado" of spanish verbs

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published