Update README.md

README.md

@@ -12,7 +12,7 @@ This model is a BERT-base-uncased model finetuned for **semantic tagging**.
 As training data, I use the English fragment (both gold and silver data) from the Parallel Meaning Bank's Universal Semantic Tags dataset [1].
 
 ## Inference
-The model is trained to make predictions for the embedded expression corresponding to the first subword of each word. Inference in the same setting as in training can be achieved with the following code ([huggingface's standard pipeline](https://huggingface.co/docs/transformers/main_classes/pipelines) does not behave as intended here). Note that the model assumes that inputs are already split into words by spaces.
+The model is trained to make predictions for the embedded representations corresponding to the first subword of each word. Inference in the same setting as in training can be achieved with the following code ([huggingface's standard pipeline](https://huggingface.co/docs/transformers/main_classes/pipelines) does not behave as intended here). Note that the model assumes that inputs are already split into words by spaces.
 ```python
 from transformers import AutoTokenizer, AutoModelForTokenClassification
 from spacy_alignments.tokenizations import get_alignments