stanza ( https://stanfordnlp.github.io/stanza/ )

.py

# %%
import stanza
import graphviz as gv

# %%
# https://qiita.com/ytaki0801/items/b8d7d4c778fe98966ce5


def p_lex(p):
    for s in '()[]{}",':
        p = p.replace(s, ' ' + s + ' ')
    return p.split()


def p_syn(p):
    t = p[0]
    del p[0]
    if t in '({["':
        r = []
        while p[0] != ')}]"'['({["'.find(t)]:
            r += [p_syn(p)]
        del p[0]
        return r
    else:
        return t


# %%
# stanza.download('ja')
nlp = stanza.Pipeline('ja', model_dir='Stanza')
doc = nlp("バラク・オバマはハワイで生まれた。 2008年に大統領に就任。")
# doc.sentences[0].print_dependencies()

# %%


def print_node(node, cnt, leafs, results):
    original_cnt = cnt
    if (len(node) == 3):
        results.append(f'\tn{cnt}'+'[label = "{'+f'{node[0]}'+'|{<pl>|<pr>}}"];')
        results.append(f"\tn{original_cnt}:pl -> n{cnt+1}")
        cnt = print_node(node[1], cnt+1, leafs, results)
        results.append(f"\tn{original_cnt}:pr -> n{cnt+1}")
        cnt = print_node(node[2], cnt+1, leafs, results)
    else:
        if (type(node[1]) == list):
            results.append(f'\tn{cnt}'+'[label = "{'+f'{node[0]}'+'|{<pl>}}"];')
            results.append(f"\tn{original_cnt}:pl -> n{cnt+1}")
            cnt = print_node(node[1], cnt+1, leafs, results)
        else:
            leafs.append(f"n{cnt}")
            results.append(f'\tn{cnt}'+'[label = "{'+f'{node[0]}|{node[1]}'+'}"];')

    return cnt


def sentence2graph(idx, s):
    root = p_syn(p_lex(str(s.constituency)))
    leafs = []
    dot = []
    dot.append("digraph graphname {\n\tnode[shape=record];")
    results = []
    print_node(root, 0, leafs, results)
    dot.extend(results)
    dot.append("\t{rank = same;"+"; ".join(leafs)+"}\n}")
    graph = gv.Source("\n".join(dot), filename=f"graph{idx}.gv", format="png")
    return graph


# %%
sentence2graph(0, doc.sentences[0])
# %%
sentence2graph(1, doc.sentences[1])

py.ipynb