Can Language Representation Models Think in Bets?