Add langchain annotation blog post (#342)

Closes #340

_drafts/langchain.md

@@ -0,0 +1,30 @@
+<!--
+
+| Scores         | Zero-shot          | Supervised          | Few-shot    |
+|----------------|-------------------:|--------------------:|-------------:
+| Micro F1-score | $$\mathbf{81.45}$$ |  $$79.88$$          | $$61.90$$   |
+| Macro F1-score | $$\mathbf{78.74}$$ |  $$77.52$$          | $$55.02$$   |
+
+
+| F1-score (per type)                      | Zero-shot           | Supervised | Few-shot  |
+|------------------------------------------|--------------------:|-----------:|-----------:
+| Supporting argument (`Argument_for`)     |  $$\mathbf{75.21}$$ | $$73.60$$  | $$48.74$$ |
+| No argument (`NoArgument`)               |  $$\mathbf{86.74}$$ | $$85.66$$  | $$72.50$$ |
+| Opposing argument (`Argument_against`)   |  $$\mathbf{74.26}$$ | $$73.30$$  | $$46.00$$ |
+
+
+-->
+
+<!--
+I have to admit that this one's definitely a **negative result.** I initially
+expected that the few-shot predictions will work better because there's added
+context from the guidelines. But it's also possible that our prompt (plus our
+sequential processing step) became a detriment to get more reliable predictions.
+
+I'm not closing my doors to this hypothesis. There's an interesting distribution
+of scores especially across category types. For example, [Levy et al.,
+(2018)](#levy2018towards) seems to perform well on `NoArgument` cases. Perhaps
+there's something in how the guideline was written that caused this? I might get
+back to this again in a more qualitative light.
+
+-->