Misc frontend changes to support sequence salience.

lit_nlp/api/layout.py

@@ -91,6 +91,7 @@ class ModuleConfig(dtypes.DataTuple):
 class LayoutSettings(dtypes.DataTuple):
   hideToolbar: bool = False
   mainHeight: int = 45
+  leftWidth: int = 50
   centerPage: bool = False
 
 

lit_nlp/client/core/modules.ts

@@ -151,6 +151,12 @@ export class LitModules extends ReactiveElement {
       (mainHeight) => {
         if (mainHeight != null) {this.upperHeight = `${mainHeight}%`;}
       });
+    this.reactImmediately(
+        () => this.modulesService.getSetting('leftWidth'), (leftWidth) => {
+          if (leftWidth != null) {
+            this.leftColumnWidth = `${leftWidth}%`;
+          }
+        });
 
     document.addEventListener('keydown', (e: KeyboardEvent) => {
       if (e.key === 'Escape') {
@@ -422,7 +428,13 @@ export class LitModules extends ReactiveElement {
     const columnSeparatorDoubleClick = (event: DragEvent) => {
       event.stopPropagation();
       event.preventDefault();
-      this.leftColumnWidth = LEFT_COLUMN_DEFAULT_WIDTH;
+      const layoutDefaultLeftWidth =
+          this.modulesService.getSetting('leftWidth');
+      if (layoutDefaultLeftWidth != null) {
+        this.leftColumnWidth = `${layoutDefaultLeftWidth}%`;
+      } else {
+        this.leftColumnWidth = LEFT_COLUMN_DEFAULT_WIDTH;
+      }
     };
 
     const leftColumnStyles = styleMap({

lit_nlp/client/elements/switch.ts

@@ -85,12 +85,17 @@ export class LitSwitch extends LitElement {
       'selected': this.selected
     });
 
+    // prettier-ignore
     return html`
       <div class=${containerClasses} @click=${toggleState}>
-        <div class='switch-label label-left'>${this.labelLeft}</div>
+        <div class='switch-label label-left'>
+          ${this.labelLeft}<slot name="labelLeft"></slot>
+        </div>
         <mwc-switch ?selected=${this.selected} ?disabled=${this.disabled}>
         </mwc-switch>
-        <div class='switch-label label-right'>${this.labelRight}</div>
+        <div class='switch-label label-right'>
+          <slot name="labelRight"></slot>${this.labelRight}
+        </div>
       </div>
     `;
   }

lit_nlp/client/elements/token_chips.css

@@ -52,4 +52,51 @@
 
 .pre-wrap {
   white-space: pre-wrap;
+}
+
+.row-break {
+  flex-basis: 100%;
+  height: 0;
+}
+
+.word-spacer {
+  width: 1em;
+}
+
+.tokens-holder-dense .word-spacer {
+  width: 0.5em;
+}
+
+/* block mode */
+.tokens-holder-display-block {
+  display: block;
+  font-size: 0; /* hack to get zero spacing between elements */
+  line-height: 22px;
+}
+
+.tokens-holder-display-block > * {
+  /* TODO: set this for all modes? */
+  font-size: 13px;  /* restore standard font size */
+}
+
+.tokens-holder-display-block .salient-token {
+  display: inline;
+  min-height: 1lh;
+  vertical-align: baseline;
+}
+
+.tokens-holder-display-block.tokens-holder-dense .salient-token span {
+   /* hack to remove extra whitespace. ugh. */
+  margin-right: -0.445ch;
+}
+
+.tokens-holder-display-block .word-spacer {
+  display: inline;
+  vertical-align: baseline;
+  white-space: pre-wrap;
+}
+
+.tokens-holder-display-block lit-tooltip {
+  --anchor-display-mode: 'inline';
+  --tooltip-position-left: 0;
 }

lit_nlp/client/elements/token_chips.ts

@@ -35,9 +35,9 @@ export interface TokenWithWeight {
   weight: number;
   selected?: boolean;
   pinned?: boolean;
-  onClick?: (e: Event) => void;
-  onMouseover?: (e: Event) => void;
-  onMouseout?: (e: Event) => void;
+  onClick?: (e: MouseEvent) => void;
+  onMouseover?: (e: MouseEvent) => void;
+  onMouseout?: (e: MouseEvent) => void;
   disableHover?: boolean;
   forceShowTooltip?: boolean;
 }
@@ -50,9 +50,33 @@ export class TokenChips extends LitElement {
   // List of tokens to display
   @property({type: Array}) tokensWithWeights: TokenWithWeight[] = [];
   @property({type: Object}) cmap: SalienceCmap = new UnsignedSalienceCmap();
-  @property({type: String})
-  tokenGroupTitle?: string;  // can be used for gradKey
+  // Group title, such as the name of the active salience method.
+  @property({type: String}) tokenGroupTitle?: string;
+  /**
+   * Dense mode, for less padding and smaller margins around each chip.
+   */
   @property({type: Boolean}) dense = false;
+  /**
+   * Block mode uses display: block and inline elements for chips, instead of
+   * a flex-row layout. This allows chips to flow across line breaks, behaving
+   * more like <span> elements and giving a much better experience for larger
+   * segments like sentences. However, this comes at the cost of more spacing
+   * artifacts and occasional issues with tooltip positioning.
+   */
+  @property({type: Boolean}) displayBlock = false;
+  /**
+   * breakNewlines removes \n at the beginning or end of a segment and inserts
+   * explicit row break elements instead. Improves readability in many settings,
+   * at the cost of "faithfulness" to the original token text.
+   */
+  @property({type: Boolean}) breakNewlines = false;
+  /**
+   * preSpace removes a leading space from a token and inserts an explicit
+   * spacer element instead. Improves readability in many settings by giving
+   * natural space between the highlight area for adjacent words, albeit at the
+   * cost of hiding where the actual spaces are in the tokenization.
+   */
+  @property({type: Boolean}) preSpace = false;
 
   static override get styles() {
     return [sharedStyles, styles];
@@ -71,17 +95,56 @@ export class TokenChips extends LitElement {
       'color': this.cmap.textCmap(tokenInfo.weight),
     });
 
-    // clang-format off
+    let tokenText = tokenInfo.token;
+
+    let preSpace = false;
+    if (this.preSpace && tokenText.startsWith(' ')) {
+      preSpace = true;
+      tokenText = tokenText.slice(1);
+    }
+
+    // TODO(b/324955623): render a gray '⏎' for newlines?
+    // Maybe make this a toggleable option, as it can be distracting.
+    // TODO(b/324955623): better rendering for multiple newlines, like \n\n\n ?
+    // Consider adding an extra ' ' on each line.
+
+    let preBreak = false;
+    let postBreak = false;
+    if (this.breakNewlines) {
+      // Logic:
+      // - \n : post-break, so blank space goes on previous line
+      // - foo\n : post-break
+      // - \nfoo : pre-break
+      // - \n\n  : pre- and post-break, shows a space on its own line
+      // - \n\n\n : pre- and post-break, two lines with only spaces
+      if (tokenText.endsWith('\n')) {
+        // Prefer post-break because this puts the blank space on the end of the
+        // previous line, rather than creating an awkward indent on the next
+        // one.
+        tokenText = tokenText.slice(0, -1) + ' ';
+        postBreak = true;
+      }
+      if (tokenText.startsWith('\n')) {
+        // Pre-break only if \n precedes some other text.
+        preBreak = true;
+        tokenText = ' ' + tokenText.slice(1);
+      }
+    }
+
+    // prettier-ignore
     return html`
+      ${preBreak ? html`<div class='row-break'></div>` : null}
+      ${preSpace ? html`<div class='word-spacer'> </div>` : null}
       <div class=${tokenClass} style=${tokenStyle} @click=${tokenInfo.onClick}
         @mouseover=${tokenInfo.onMouseover} @mouseout=${tokenInfo.onMouseout}>
         <lit-tooltip content=${tokenInfo.weight.toPrecision(3)}
           ?forceShow=${Boolean(tokenInfo.forceShowTooltip)}
           ?disabled=${Boolean(tokenInfo.disableHover)}>
-          <span class='pre-wrap' slot="tooltip-anchor">${tokenInfo.token}</span>
+          <span class='pre-wrap' slot="tooltip-anchor">${tokenText}</span>
         </lit-tooltip>
-      </div>`;
-    // clang-format on
+      </div>
+      ${postBreak ? html`<div class='row-break'></div>` : null}
+      `;
   }
 
   override render() {
@@ -92,17 +155,17 @@ export class TokenChips extends LitElement {
     const holderClass = classMap({
       'tokens-holder': true,
       'tokens-holder-dense': this.dense,
+      'tokens-holder-display-block': this.displayBlock,
     });
 
-    // clang-format off
+    // prettier-ignore
     return html`
       <div class="tokens-group">
         ${this.tokenGroupTitle ? this.tokenGroupTitle : ''}
         <div class=${holderClass}>
           ${tokensDOM}
         </div>
       </div>`;
-    // clang-format on
   }
 }
 

lit_nlp/client/elements/token_chips_test.ts

@@ -31,7 +31,21 @@ const TESTDATA: Array<{tokensWithWeights: TokenWithWeight[]}> = [
       {token: 'hello', weight: 0.7, selected: true, pinned: true},
       {token: 'world', weight: 0.3}
     ],
-  }
+  },
+  {
+    // for testing preSpace mode
+    tokensWithWeights: [
+      {token: 'foo', weight: 0.7, selected: true, pinned: true},
+      {token: ' bar', weight: 0.3}, {token: 'baz', weight: 0.5}
+    ],
+  },
+  {
+    // for testing breakNewlines mode
+    tokensWithWeights: [
+      {token: 'foo', weight: 0.7}, {token: '\nbar', weight: 0.3},
+      {token: '\n\n', weight: 0.1}, {token: 'baz\n', weight: 0.5}
+    ],
+  },
 ];
 
 describe('token chips test', () => {
@@ -60,6 +74,58 @@ describe('token chips test', () => {
       expect(tokenElements[0].children[0]).toBeInstanceOf(LitTooltip);
     });
 
+    it('should break spaces in preSpace mode', async () => {
+      tokenChips.preSpace = true;
+      await tokenChips.updateComplete;
+
+      const tokenElements =
+          tokenChips.renderRoot.querySelectorAll<HTMLDivElement>(
+              'div.salient-token');
+      expect(tokenElements.length).toEqual(tokensWithWeights.length);
+      for (let i = 0; i < tokenElements.length; i++) {
+        const elem = tokenElements[i];
+        const expectedToken = tokensWithWeights[i].token;
+        if (expectedToken.startsWith(' ')) {
+          // Space moved to a word spacer.
+          expect(elem.innerText).toEqual(expectedToken.slice(1));
+          expect(elem.previousElementSibling?.classList ?? [])
+              .toContain('word-spacer');
+        } else {
+          // Space intact, no word spacer.
+          expect(elem.innerText).toEqual(expectedToken);
+          if (i > 0) {
+            expect(elem.previousElementSibling?.classList ?? [])
+                .toContain('salient-token');
+          }
+        }
+      }
+    });
+
+    it('should break newlines in breakNewlines mode', async () => {
+      tokenChips.breakNewlines = true;
+      await tokenChips.updateComplete;
+
+      const tokenElements =
+          tokenChips.renderRoot.querySelectorAll<HTMLDivElement>(
+              'div.salient-token');
+      expect(tokenElements.length).toEqual(tokensWithWeights.length);
+      for (let i = 0; i < tokenElements.length; i++) {
+        const elem = tokenElements[i];
+        let expectedToken = tokensWithWeights[i].token;
+        if (expectedToken.endsWith('\n')) {
+          expectedToken = expectedToken.slice(0, -1) + ' ';
+          expect(elem.nextElementSibling?.classList ?? [])
+              .toContain('row-break');
+        }
+        if (expectedToken.startsWith('\n')) {
+          expectedToken = ' ' + expectedToken.slice(1);
+          expect(elem.previousElementSibling?.classList ?? [])
+              .toContain('row-break');
+        }
+        expect(elem.innerText).toEqual(expectedToken);
+      }
+    });
+
     it('should mark a selected token', async () => {
       const tokenElements =
           tokenChips.renderRoot.querySelectorAll<HTMLDivElement>(

lit_nlp/client/lib/types.ts

@@ -410,8 +410,10 @@ export declare interface LitCanonicalLayout {
  */
 export declare interface LayoutSettings {
   hideToolbar?: boolean;
-  /** The default height of #upper-right, as a percentage of the parent. */
+  /** The default height of the 'upper' section, as a percentage. */
   mainHeight?: number;
+  /** The default width of the  'left' section, as a percentage. */
+  leftWidth?: number;
   centerPage?: boolean;
 }