Seol mar théacs é seo: Robust processing of spoken situated dialogue