Word analysis¶

An analysis is an Analysis object returned by the analyze and analyze_many methods of the analyzer: the lemma, the part of speech and the Universal Dependencies features, the probability, the words of the token and the source of the analysis. Analyses are immutable; they can be compared (the probability takes no part in the comparison, so an analysis equals its form from forms()), put into sets and pickled. The forms of a word are given by the forms and inflect methods - see Inflection.

CodeResult

from esmorph import Analyzer

morph = Analyzer()
analysis = morph.analyze("cantaban")[0]
analysis.word, analysis.lemma, analysis.pos, analysis.feats, analysis.words, analysis.source

('cantaban',
 'cantar',
 'VERB',
 {'Mood': 'Ind', 'Number': 'Plur', 'Person': '3', 'Tense': 'Imp', 'VerbForm': 'Fin'},
 ('cantaban',),
 'dictionary')

Attributes¶

Attribute	Type	Description
`word`	`str`	The word form as the dictionary spells it, with its accents; a verb with clitics as written, in lower case; a token of the rules as written, without invisible characters, with combining marks joined to their letters and with `-` for the hyphens U+2010 and U+2011
`lemma`	`str`	The lemma
`pos`	`str`	UD part of speech: `NOUN`, `VERB`, `ADJ`, `DET`, `ADP` and so on
`feats`	`dict[str, str]`	UD features: `{"Gender": "Fem", "Number": "Plur"}`; a new dictionary each time, changing it does not change the analysis
`prob`	`float`	Share of the analysis among the analyses of the word; it sums to 1 over all of them
`words`	`tuple[str, ...]`	Words of the token: a verb and its clitics, the words of a contraction (`del`, `al`, the older `desta`), or the word itself; the analysis describes the first of them
`source`	`str`	Where the analysis comes from: `"dictionary"` - the word form is in the dictionary, `"clitics"` - a verb of the dictionary with clitics, `"rule"` - an analysis by rule, `"guess"` - a guessed analysis of an unknown word

Parts of speech and features¶

Parts of speech, features and lemmas follow the Universal Dependencies guidelines and, where they are silent, the conventions of the largest Spanish treebank, UD Spanish AnCora, so the output can be compared with spaCy, Stanza and other UD tools:

determiners, pronouns, prepositions and conjunctions have the lemmas and features of AnCora: la is the article el (Definite=Def, PronType=Art) or the pronoun él, se is él with Reflex=Yes, su is DET with Poss=Yes, muy is a form of mucho, and so are their older and spoken spellings, as variants: usté is a form of tú; a demonstrative pronoun with an accent (éste) is a variant of the one without;
ser, estar, haber, poder, deber, soler, querer and saber have both AUX and VERB analyses; hay is first of all the verb haber;
a participle is a verb with the infinitive as the lemma (VerbForm=Part, Tense=Past) and an adjective with the masculine singular as the lemma (VerbForm=Part without Tense);
verbs have no -se in the lemma: quejarse is quejar with the clitic se;
the conditional and the imperative have no Tense; a proper name has no features, and its lemma has a capital letter: madrid and Madrid are Madrid, PSOE stays PSOE;
an adjective in -ísimo is its own lemma with Degree=Abs.

If the part of speech of a word depends on context, the word has both analyses: que is PRON and SCONJ, la is DET and PRON, a participle is VERB and ADJ. Features: gender, number, person, tense, mood, verb form, degree, negation (Polarity=Neg of no), definiteness (Definite), the type of a pronoun or a determiner (PronType), possessiveness (Poss), reflexivity (Reflex), the case of personal pronouns (Case) and whether they follow a preposition (PrepCase), politeness (Polite=Form of usted), the type of a numeral (NumType) and how it is written (NumForm), and the kind of a sign of punctuation (PunctType, PunctSide).

Order of analyses¶

Analyses go from the most probable to the least. prob is the share of an analysis among the analyses of the word, by how often the treebanks UD Spanish AnCora and GSD have it; for a word they have not seen, the shares are equal and the order is a guess by the ending of the word and the most common features, and a word goes before its older spelling with other accents: á is first of all a, tambien is también.

CodeResult

[(a.lemma, a.pos, round(a.prob, 2)) for a in morph.analyze("casa")][:2]

[('casa', 'NOUN', 0.99), ('casar', 'VERB', 0.01)]

Words with clitics¶

A verb with enclitic pronouns is analyzed as its verb: the analysis describes the infinitive, the gerund or the imperative, and words gives the words of the token, as Universal Dependencies has them. In older and literary writing a finite form takes enclitics too, when the accent of the token marks it: díjole is dijo + le. del and al also have their two words, as do the contractions of de with a pronoun or a demonstrative in older writing: desta is de + esta. A word of the dictionary is not split: comete is a form of cometer, though a few are mostly verbs with clitics: verse is first of all ver + se.

CodeResult

[(a.lemma, a.pos, a.words, a.source) for a in morph.analyze("dámelo")]

[('dar', 'VERB', ('da', 'me', 'lo'), 'clitics')]

Spelling¶

Case does not matter for finding a word in the dictionary, though a token with a capital letter may also be a name. A word spelled as in the dictionary gives the analyses of that spelling: publico is not público. A word absent as written may be a verb with clitics (verte is ver + te, not verté); otherwise it is looked for without its accents and with a lost diaeresis, and word shows the spelling of the dictionary; for a verb with clitics word is the token as written, in lower case.

CodeResult

[morph.analyze(word)[0].word for word in ["CAMIÓN", "camion", "pinguino"]]

['camión', 'camión', 'pingüino']

Tokens out of the dictionary¶

A token that the dictionary does not have gets the analyses of rules, with source equal to "rule" and word as written:

a sign of punctuation is PUNCT with the PunctType of its signs if they are of one kind, and PunctSide for brackets and question and exclamation marks; a symbol such as %, € or ² is SYM;
a number in digits is NUM with NumForm=Digit|NumType=Card, a year too, also with a sign, a fraction, a power or spaces between thousands (-5, 37½, 10², 10 000); the lemma drops the periods between thousands (10.000 is 10000) and a unit after the number (500cc is 500);
an ordinal number in digits is ADJ with NumForm=Digit|NumType=Ord, its gender and number, and the masculine singular as the lemma: 2.ª is 2.º;
a Roman numeral is NUM with NumForm=Roman, less often a name; in lower case it is a numeral only: xii;
a token with a capital letter is a name, PROPN, with itself as the lemma: Josep, CiU, W., TV-3;
a link, an email, a hashtag or a mention in lower case is SYM: www.aeat.es, #hola;
an abbreviation with a period in lower case that the dictionary does not have is a noun with itself as the lemma (fig.), and so is a lowercase letter (b).

A lowercase abbreviation with a period that the dictionary has without the period gets the nouns and the symbols of that word, with source equal to "dictionary": kg. is kg. A word of the dictionary with a period of another part of speech gets no analyses: come. is a verb with the period of a sentence.

A word with a capital letter that the dictionary knows may be a name too, with its share. An initial (D.) and a word without accents that the dictionary has only with them (Angel, Perez) are mostly names, so their name goes first, while another word with a capital letter is a name less often (Estados, Comisión), so its name usually goes after the analyses of the dictionary. Of the words in capitals, a word of up to four letters is mostly an acronym (ETA, PP), so its name usually goes first, while a longer word and a function word - a preposition, a determiner, a pronoun, a conjunction, an auxiliary - are mostly words of a headline (CAMIONES, EL), so their name usually goes after the analyses of the dictionary. A Roman numeral that the dictionary knows as a word, such as VI, XI or X, is a numeral first. A token with a capital letter that splits as a verb with clitics may be a name too: an imperative without an accent mostly is one (Mandela is first of all PROPN), while an infinitive, a gerund, an imperative with an accent and a finite form mostly are verbs (Hacerlo, Déjame, Levantóse). A lowercase word is a name rarely: miró is first of all mirar, Miró a name. An analysis of a rule has no other forms: forms() gives the analysis itself. is_known tells only the words of the dictionary.

CodeResult

[(a.lemma, a.pos, a.feats, a.source) for a in morph.analyze("10.000")]
[(a.lemma, a.pos, round(a.prob, 2)) for a in morph.analyze("ETA")]

[('10000', 'NUM', {'NumForm': 'Digit', 'NumType': 'Card'}, 'rule')]
[('ETA', 'PROPN', 0.82), ('eta', 'NOUN', 0.18)]

Unknown words¶

A lowercase word that neither the dictionary nor the rules know gets guessed analyses, with source equal to "guess":

a word with a hyphen is analyzed by its last part of an open class or a numeral: palestino-israelíes is a form of palestino-israelí;
a word with a known prefix is analyzed by the dictionary word without it, of four letters at least after a prefix of three: supergallos is a form of supergallo, sobreandan of sobreandar;
a verb with clitics is analyzed by its guessed gerund, or by its guessed infinitive before se or a clitic of the third person: autogenerarse is autogenerar + se;
any other word is analyzed by the dictionary words with the same ending of up to five letters: their parts of speech, features and inflection, by how many words have them and how often AnCora train has their features; a word in -ísimo is an adjective of Degree=Abs: esbeltísima is a form of esbeltísimo.

A guessed word has its forms, guessed too: forms() and inflect() work as for a word of the dictionary. is_known stays false for it.

CodeResult

[(a.lemma, a.pos, round(a.prob, 2)) for a in morph.analyze("zumbeábamos")]
morph.analyze("zumbeábamos")[0].inflect(Tense="Pres", Person="1", Number="Sing")

[('zumbear', 'VERB', 0.9), ('zumbeábar', 'VERB', 0.09), ('zumbeábar', 'VERB', 0.01)]
'zumbeo'