Abstract: Chord symbols, typically notating the root note and the chord quality, are extensively used yet oversimplified representation of tonal harmony and chord progressions in popular music. In spite of its convenience, the chord symbol notation only provides basic information about the chordal configuration, and leaves much room for interpretation. With such limitations, an algorithm generating merely chord symbols is usually insufficient for a wide range of music genres such as jazz. To solve this problem, we propose chord jazzification, a process to generate realistic chord configurations in jazz style. With deep learning approaches, we decompose chord jazzification into coloring and voicing. Coloring concerns the choice of color tones, while voicing concerns the configurations of chords. We also create a new dataset featuring interpretations of chord symbols in pop-jazz compositions. By conducting experiments on the new dataset, we show that 1) the two-stage process outperforms an end-to-end generation approach in modeling chord configurations, and 2) attention-based models are better at capturing the structure of chord sequences in comparison with recurrent neural networks.