Studies of variation in continuous speech converge towards the conclusion that in everyday speech, words are often produced with reduced variants: some segments are shortened or completely absent. We describe an initiative to automatically exploit spoken corpora, in order to better understand linguistic behavior in spontaneous speech. This study focuses on the reduction of the postposed definite article in Romanian. The Romanian corpora used here cover several speaking styles including both prepared and spontaneous speech, such as broadcast news and debates, elicited dialogues and monologues on suggested topics. Taken together, the data sets contain more than 10 hours of speech produced in a variety of communicative frameworks. The deletion of the definite article -l, i.e. L-dropping in continuous speech, is investigated across speaking styles using pronunciation variants aligned with the speech. The main question addressed in the study is the influence of speaking style on the distribution of L-dropping. We examine the role of the surrounding context in L-dropping and L-retention. The results show that, in prepared speech and broadcast news, deletion is triggered by the context as a consequence of the communicative framework and the nature of the following segment (following vowel-initial words favor L-retention, while consonant-initial words favor L-dropping). In spontaneous speech, L-dropping is more frequent and the context is less important in predicting the occurrence of reduced variants than in other speaking styles.