Making inflected forms recognized with the verb and adjective dictionaries
If you put a coined verb such as 카톡하다 or an adjective such as 신박하다 in the noun dictionary, it is recognized only when the text is exactly
"카톡하다." But Korean predicates conjugate endlessly: "카톡했다, 카톡하는, 카톡하면." If you register it in Bareun's verb dictionary (vv_set) or adjective dictionary (va_set),
one headword makes its inflected forms recognized automatically too.
This document explains how the verb and adjective dictionaries work, how to register entries, and how inflected forms are analyzed.
Why a predicate dictionary rather than the noun dictionary
흘낏하다 is a verb derived from the adverb 흘낏. In training data, such words are often
tagged awkwardly as "adverb + verb-deriving suffix."
Bareun keeps a separate verb dictionary so that such words are properly tagged as verbs.
Adjectives such as 소녀스럽다 are managed in the adjective dictionary in the same way.
Users can also define and register new predicates, and what goes in is the base form (dictionary form).
- Verbs: base forms ending in
~하다/~거리다/~대다, such as카톡하다and맑내하다 - Adjectives: base forms ending in
~하다/~스럽다, such as로맨틱하다,신박하다, and판타스틱하다
How automatic recognition of inflected forms works
When you register a base form in the verb or adjective dictionary, Bareun separates the stem during analysis and recognizes
inflected forms where it combines with endings. That is, if you register 카톡하다, the stem 카톡하 combined with final endings, connective endings, adnominal endings,
and so on, such as "카톡했다," "카톡하는," and "카톡하며," are all analyzed as the same verb.
Registration example
This is an example of filling the verb and adjective dictionaries with the bareunpy client.
from bareunpy import Tagger
my_tagger = Tagger('YOUR-API-KEY', 'localhost', port=5656)
cust_dic = my_tagger.custom_dict("trend_words")
# Verb dictionary: register as base forms (~하다)
cust_dic.copy_vv_set({'카톡하다', '인스타하다', '맑내하다'})
# Adjective dictionary: register as base forms (~하다/~스럽다)
cust_dic.copy_va_set({'신박하다', '판타스틱하다', '소녀스럽다'})
cust_dic.update()
You do not need to register inflected forms. One base form is enough.
What the analyzed inflected forms look like
After registration, when you analyze a sentence with inflected forms, the stem is separated as a verb or adjective and the following ending as a separate morpheme.
어제 친구랑 카톡하다가 신박한 아이디어가 떠올랐어요.
카톡하/VV, IN_CUSTOM_DICT
다가/EC
신박하/VA, IN_CUSTOM_DICT
ㄴ/ETM
Even though only the base form 카톡하다 was registered, the inflected form "카톡하다가" is split into 카톡하/VV + 다가/EC,
and for 신박하다, the adnominal form "신박한" is analyzed as 신박하/VA + ㄴ/ETM.
Both stems have out_of_vocab marked as IN_CUSTOM_DICT, so you can tell they came from the custom dictionary.
Strength of the predicate dictionary
If you register it as a noun, it matches only the string "카톡하다," but if you put it in the verb dictionary, a single headword can cover an unlimited number of inflected forms. The effect of the predicate dictionary is greater for newly coined predicates.
Register base forms
Do not register inflected forms (카톡했다) or bare stems (카톡하). The verb and adjective dictionaries are designed to accept
base forms ending in ~다. Also, entries cannot contain spaces or some symbols.
Frequently asked questions
Q. Do I need to put inflected forms in the verb dictionary?
No. Register just the base form (카톡하다). Bareun separates the stem and automatically recognizes
inflected forms combined with various endings.
Q. I can't tell whether a ~하다 word is a verb or an adjective. Where do I put it?
If it describes an action of the subject, it is a verb (vv_set); if it describes a state or property, it is an adjective (va_set).
For example, "카톡하다" is an action, so a verb, and "신박하다" is a state, so an adjective.
If you put it in the wrong dictionary, the POS of its inflected forms comes out differently from what you intended.
Q. Can I put the same word in both the verb and adjective dictionaries?
Putting the same headword in both predicate dictionaries at once is reported as a conflict. If its usage cannot be settled as one, register it only in the more frequently used POS.