Using computational linguistic analysis of intraday firm-level news releases, this study models the relation between public information flows and stock volatility under different regimes. We analyze how the hourly return volatility of S&P100 stocks from 2000 to 2010 are linked to the various linguistics-based sentiment scores of the news releases, which are obtained from the RavenPack News Analytics Database. Results from the Markov Regime-Switching GARCH (MRS-GARCH) model indicate that firm-specific news sentiment is more significant in quantifying intraday volatility persistence in the calm (low-volatility) state than the turbulent (high-volatility) state. Furthermore, the impact of news sentiment differs across industries and firm size.