It would complicate your parsing a bit, but I think you could have an even simpler format by omitting an extra layer of array for sibling nodes (just use them for children). The structure translation would still be unambiguous.

[ family => { name => 'Kawasaki' },
  [
    father => 'Yasushisa' ,
    mother => { hair => 'short'} => 'Chizuko' ,
    children =>
    [
      girl => 'Shiori',
      boy  => { hair => 'black' } => 'Yasuke',
      boy  => 'Kairi'
    ]
  ]
]
[download]

<family name="Kawasaki">
    <father>Yasuhisa</father>
    <mother hair="short">Chizuko</mother>
    <children>
        <girl>Shiori</girl>
        <boy hair="black">Yusuke</boy>
        <boy>Kairi</boy>
    </children>
</family>
[download]